From 33b4f1a55cc0fc6b8be138d5097cf7ad8ec487fc Mon Sep 17 00:00:00 2001
From: Ligeng Zhu <Lyken17@users.noreply.github.com>
Date: Tue, 7 Jan 2025 10:48:15 +0800
Subject: [PATCH] Cosmos Nemotron release

---
 data_prepare/coyo/coyo_downloader.py                            | 2 ++
 data_prepare/coyo/coyo_splitter.py                              | 2 ++
 data_prepare/mmc4/mmc4_downloader.py                            | 2 ++
 data_prepare/mmc4/mmc4_filter_and_counter.py                    | 2 ++
 data_prepare/mmc4/mmc4_merger.py                                | 2 ++
 data_prepare/panda_split.py                                     | 2 ++
 data_prepare/sft/ART1_2.py                                      | 2 ++
 data_prepare/sft/ESTVQA.py                                      | 2 ++
 data_prepare/sft/LSVT.py                                        | 2 ++
 data_prepare/sft/POIE.py                                        | 2 ++
 data_prepare/sft/ReCTS.py                                       | 2 ++
 data_prepare/sft/SROIE.py                                       | 2 ++
 data_prepare/sft/merge_idefics2.py                              | 2 ++
 data_prepare/sft/merge_llava_onevision.py                       | 2 ++
 data_prepare/sft/merge_llava_onevision_eagle.py                 | 2 ++
 data_prepare/sft/mtwi.py                                        | 2 ++
 data_prepare/sft/preprocess_art_shangy.py                       | 2 ++
 data_prepare/sft/preprocess_cambrian.py                         | 2 ++
 data_prepare/sft/preprocess_cambrian_eagle.py                   | 2 ++
 data_prepare/sft/preprocess_docreason.py                        | 2 ++
 data_prepare/sft/preprocess_flan.py                             | 2 ++
 data_prepare/sft/preprocess_idefics2.py                         | 2 ++
 data_prepare/sft/preprocess_idefics2_eagle.py                   | 2 ++
 data_prepare/sft/preprocess_kvqa.py                             | 2 ++
 data_prepare/sft/preprocess_llava_onevision.py                  | 2 ++
 data_prepare/sft/preprocess_m3it.py                             | 2 ++
 data_prepare/sft/preprocess_metamathqa.py                       | 2 ++
 data_prepare/sft/preprocess_viquae.py                           | 2 ++
 data_prepare/sft/split_vflan.py                                 | 2 ++
 data_prepare/sft/unichart_pretrain.py                           | 2 ++
 data_prepare/sft/unichart_sft.py                                | 2 ++
 llava/__init__.py                                               | 2 ++
 llava/cli/eval.py                                               | 2 ++
 llava/cli/infer.py                                              | 2 ++
 llava/cli/run.py                                                | 2 ++
 llava/cli/upload2hf.py                                          | 2 ++
 llava/constants.py                                              | 2 ++
 llava/conversation.py                                           | 2 ++
 llava/data/__init__.py                                          | 2 ++
 llava/data/base.py                                              | 2 ++
 llava/data/builder.py                                           | 2 ++
 llava/data/collate.py                                           | 2 ++
 llava/data/dataset.py                                           | 2 ++
 llava/data/dataset_impl/__init__.py                             | 2 ++
 llava/data/dataset_impl/coyo_qa.py                              | 2 ++
 llava/data/dataset_impl/coyo_recap.py                           | 2 ++
 llava/data/dataset_impl/dummy.py                                | 2 ++
 llava/data/dataset_impl/lita.py                                 | 2 ++
 llava/data/dataset_impl/llava.py                                | 2 ++
 llava/data/dataset_impl/llava_cot.py                            | 2 ++
 llava/data/dataset_impl/panda70m.py                             | 2 ++
 llava/data/dataset_impl/sam.py                                  | 2 ++
 llava/data/dataset_impl/textocr.py                              | 2 ++
 llava/data/dataset_impl/utils.py                                | 2 ++
 llava/data/datasets_mixture.py                                  | 2 ++
 llava/data/simple_vila_webdataset.py                            | 2 ++
 llava/data/utils.py                                             | 2 ++
 llava/entry.py                                                  | 2 ++
 llava/eval/__init__.py                                          | 2 ++
 llava/eval/cinepile.py                                          | 2 ++
 llava/eval/egoschema.py                                         | 2 ++
 llava/eval/eval_refcoco.py                                      | 2 ++
 llava/eval/eventbench.py                                        | 2 ++
 llava/eval/lmms/models/__init__.py                              | 2 ++
 llava/eval/lmms/models/vila_internal.py                         | 2 ++
 llava/eval/lmms/tasks/__init__.py                               | 2 ++
 llava/eval/lmms/tasks/videomme.py                               | 2 ++
 llava/eval/m4c_evaluator.py                                     | 2 ++
 llava/eval/mathvista.py                                         | 2 ++
 llava/eval/mathvista_utils/calculate_score.py                   | 2 ++
 llava/eval/mathvista_utils/extract_answer.py                    | 2 ++
 llava/eval/mathvista_utils/prompts/ext_ans.py                   | 2 ++
 llava/eval/mathvista_utils/utilities.py                         | 2 ++
 llava/eval/mmmu_utils/eval_utils.py                             | 2 ++
 llava/eval/model_refcoco.py                                     | 2 ++
 llava/eval/model_vqa_video.py                                   | 2 ++
 llava/eval/model_vqa_videodemo.py                               | 2 ++
 llava/eval/rtl.py                                               | 2 ++
 llava/eval/textvqa.py                                           | 2 ++
 llava/eval/video/eval_benchmark_1_correctness.py                | 2 ++
 llava/eval/video/eval_benchmark_2_detailed_orientation.py       | 2 ++
 llava/eval/video/eval_benchmark_3_context.py                    | 2 ++
 llava/eval/video/eval_benchmark_4_temporal.py                   | 2 ++
 llava/eval/video/eval_benchmark_5_consistency.py                | 2 ++
 llava/eval/video/eval_video_qa.py                               | 2 ++
 llava/eval/video/model_vqa_videodemo_benchmark.py               | 2 ++
 llava/eval/video/utils.py                                       | 2 ++
 llava/eval/vision_niah_vila/eval_vision_niah.py                 | 2 ++
 llava/eval/vision_niah_vila/produce_haystack_embedding.py       | 2 ++
 llava/eval/vision_niah_vila/produce_needle_embedding.py         | 2 ++
 llava/eval/vision_niah_vila/zigzag_ring_attn/modeling_qwen2.py  | 2 ++
 llava/eval/vision_niah_vila/zigzag_ring_attn/monkey_patch.py    | 2 ++
 llava/eval/vision_niah_vila/zigzag_ring_attn/prepare_inputs.py  | 2 ++
 llava/eval/vnbench.py                                           | 2 ++
 llava/media.py                                                  | 2 ++
 llava/mm_utils.py                                               | 2 ++
 llava/model/FloatPointQuantizeTorch.py                          | 2 ++
 llava/model/FloatPointQuantizeTriton.py                         | 2 ++
 llava/model/__init__.py                                         | 2 ++
 llava/model/apply_delta.py                                      | 2 ++
 llava/model/builder.py                                          | 2 ++
 llava/model/coat/activation/__init__.py                         | 2 ++
 .../activation/fake_quantization/FloatPointQuantizeTorch.py     | 2 ++
 .../activation/fake_quantization/FloatPointQuantizeTriton.py    | 2 ++
 .../coat/activation/fake_quantization/quantize_function.py      | 2 ++
 llava/model/coat/activation/fake_quantization/utils.py          | 2 ++
 llava/model/coat/activation/models/_fp8_quantization_config.py  | 2 ++
 llava/model/coat/activation/models/_fp8_weightcache.py          | 2 ++
 llava/model/coat/activation/models/_fp8manager.py               | 2 ++
 llava/model/coat/activation/models/coat_llama.py                | 2 ++
 .../model/coat/activation/models/coat_llama_convert_from_hf.py  | 2 ++
 llava/model/coat/activation/models/coat_olmo.py                 | 2 ++
 llava/model/coat/activation/real_quantization/__init__.py       | 2 ++
 llava/model/coat/activation/real_quantization/_dequantize.py    | 2 ++
 llava/model/coat/activation/real_quantization/_division.py      | 2 ++
 .../coat/activation/real_quantization/_division_transpose.py    | 2 ++
 llava/model/coat/activation/real_quantization/_memory_io.py     | 2 ++
 llava/model/coat/activation/real_quantization/_quantize.py      | 2 ++
 .../coat/activation/real_quantization/_quantize_pertensor.py    | 2 ++
 .../real_quantization/_quantize_pertensor_transpose.py          | 2 ++
 llava/model/coat/activation/real_quantization/_transpose.py     | 2 ++
 llava/model/coat/activation/real_quantization/add_bwd.py        | 2 ++
 llava/model/coat/activation/real_quantization/add_fwd.py        | 2 ++
 llava/model/coat/activation/real_quantization/common.py         | 2 ++
 llava/model/coat/activation/real_quantization/fp8linear.py      | 2 ++
 .../coat/activation/real_quantization/func_layernorm_noparam.py | 2 ++
 llava/model/coat/activation/real_quantization/func_quantize.py  | 2 ++
 llava/model/coat/activation/real_quantization/func_rmsnorm.py   | 2 ++
 llava/model/coat/activation/real_quantization/gelu_bwd.py       | 2 ++
 .../model/coat/activation/real_quantization/gelu_bwd_legacy.py  | 2 ++
 llava/model/coat/activation/real_quantization/gelu_fwd.py       | 2 ++
 llava/model/coat/activation/real_quantization/linear.py         | 2 ++
 llava/model/coat/activation/real_quantization/mul_bwd.py        | 2 ++
 llava/model/coat/activation/real_quantization/mul_bwd_legacy.py | 2 ++
 .../model/coat/activation/real_quantization/mul_bwd_silu_fwd.py | 2 ++
 llava/model/coat/activation/real_quantization/mul_fwd.py        | 2 ++
 llava/model/coat/activation/real_quantization/silu_bwd.py       | 2 ++
 .../model/coat/activation/real_quantization/silu_bwd_legacy.py  | 2 ++
 llava/model/coat/activation/real_quantization/silu_fwd.py       | 2 ++
 llava/model/coat/activation/utils.py                            | 2 ++
 llava/model/coat/fp8_trainer.py                                 | 2 ++
 llava/model/coat/optimizer/fp8_adamw.py                         | 2 ++
 llava/model/coat/optimizer/kernels/setup.py                     | 2 ++
 llava/model/configuration_llava.py                              | 2 ++
 llava/model/consolidate.py                                      | 2 ++
 llava/model/encoders/__init__.py                                | 2 ++
 llava/model/encoders/base.py                                    | 2 ++
 llava/model/encoders/image/__init__.py                          | 2 ++
 llava/model/encoders/image/basic.py                             | 2 ++
 llava/model/encoders/video/__init__.py                          | 2 ++
 llava/model/encoders/video/basic.py                             | 2 ++
 llava/model/encoders/video/tsp.py                               | 2 ++
 llava/model/language_model/builder.py                           | 2 ++
 llava/model/language_model/configuration_quantize.py            | 2 ++
 llava/model/language_model/fp8_qwen2_convert_from_hf.py         | 2 ++
 llava/model/language_model/fp8activationqwen2.py                | 2 ++
 llava/model/language_model/fp8activationresidualqwen2.py        | 2 ++
 llava/model/language_model/fp8linearqwen2.py                    | 2 ++
 llava/model/language_model/llava_llama.py                       | 2 ++
 llava/model/language_model/qllama.py                            | 2 ++
 llava/model/language_model/qllava_qllama.py                     | 2 ++
 llava/model/language_model/qmemllama.py                         | 2 ++
 llava/model/language_model/realqmemllama.py                     | 2 ++
 llava/model/liger/cross_entropy.py                              | 2 ++
 llava/model/liger/utils.py                                      | 2 ++
 llava/model/llava_arch.py                                       | 2 ++
 llava/model/loss.py                                             | 2 ++
 llava/model/make_delta.py                                       | 2 ++
 llava/model/multimodal_encoder/builder.py                       | 2 ++
 llava/model/multimodal_encoder/clip_encoder.py                  | 2 ++
 llava/model/multimodal_encoder/image_processor.py               | 2 ++
 .../model/multimodal_encoder/intern/configuration_intern_vit.py | 2 ++
 llava/model/multimodal_encoder/intern/flash_attention.py        | 2 ++
 llava/model/multimodal_encoder/intern/modeling_intern_vit.py    | 2 ++
 llava/model/multimodal_encoder/intern_encoder.py                | 2 ++
 llava/model/multimodal_encoder/radio_encoder.py                 | 2 ++
 llava/model/multimodal_encoder/radio_torchhub_encoder.py        | 2 ++
 llava/model/multimodal_encoder/siglip/__init__.py               | 2 ++
 llava/model/multimodal_encoder/siglip/modeling_siglip.py        | 2 ++
 llava/model/multimodal_encoder/siglip_encoder.py                | 2 ++
 llava/model/multimodal_encoder/vision_encoder.py                | 2 ++
 llava/model/multimodal_encoder/visualize_features.py            | 2 ++
 llava/model/multimodal_projector/base_projector.py              | 2 ++
 llava/model/multimodal_projector/builder.py                     | 2 ++
 llava/model/qfunction.py                                        | 2 ++
 llava/model/qlinear_te.py                                       | 2 ++
 llava/model/quantization/FloatPointQuantizeTorch.py             | 2 ++
 llava/model/quantization/FloatPointQuantizeTriton.py            | 2 ++
 llava/model/quantization/QAct.py                                | 2 ++
 llava/model/quantization/QAdd.py                                | 2 ++
 llava/model/quantization/QFunction.py                           | 2 ++
 llava/model/quantization/QGELU.py                               | 2 ++
 llava/model/quantization/QIdentity.py                           | 2 ++
 llava/model/quantization/QLayerNorm.py                          | 2 ++
 llava/model/quantization/QLinear.py                             | 2 ++
 llava/model/quantization/QMul.py                                | 2 ++
 llava/model/quantization/Qconfig.py                             | 2 ++
 llava/model/quantization/__init__.py                            | 2 ++
 llava/model/quantization/utils.py                               | 2 ++
 llava/model/qutils.py                                           | 2 ++
 llava/model/realquantize/common.py                              | 2 ++
 llava/model/realquantize/division.py                            | 2 ++
 llava/model/realquantize/division_transpose.py                  | 2 ++
 llava/model/realquantize/linear.py                              | 2 ++
 llava/model/realquantize/quantize_and_transpose.py              | 2 ++
 llava/model/realquantize/trans_grad_bias.py                     | 2 ++
 llava/model/utils/__init__.py                                   | 2 ++
 llava/model/utils/packing.py                                    | 2 ++
 llava/model/utils/utils.py                                      | 2 ++
 llava/train/__init__.py                                         | 2 ++
 llava/train/args.py                                             | 2 ++
 llava/train/callbacks/autoresume_callback.py                    | 2 ++
 llava/train/deepspeed_replace/runtime/zero/mics.py              | 2 ++
 llava/train/llava_trainer.py                                    | 2 ++
 llava/train/sequence_parallel/__init__.py                       | 2 ++
 llava/train/sequence_parallel/all_to_all.py                     | 2 ++
 llava/train/sequence_parallel/globals.py                        | 2 ++
 llava/train/sequence_parallel/hybrid_attn.py                    | 2 ++
 llava/train/sequence_parallel/input_utils.py                    | 2 ++
 llava/train/sequence_parallel/monkey_patch.py                   | 2 ++
 llava/train/sequence_parallel/ring/__init__.py                  | 2 ++
 llava/train/sequence_parallel/ring/ring_flash_attn.py           | 2 ++
 llava/train/sequence_parallel/ring/ring_flash_attn_varlen.py    | 2 ++
 llava/train/sequence_parallel/ring/stripe_flash_attn.py         | 2 ++
 llava/train/sequence_parallel/ring/triton_utils.py              | 2 ++
 llava/train/sequence_parallel/ring/utils.py                     | 2 ++
 llava/train/sequence_parallel/ring/zigzag_ring_flash_attn.py    | 2 ++
 .../sequence_parallel/ring/zigzag_ring_flash_attn_varlen.py     | 2 ++
 llava/train/sequence_parallel/ulysses_attn.py                   | 2 ++
 llava/train/slurm_utils.py                                      | 2 ++
 llava/train/train.py                                            | 2 ++
 llava/train/train_hybrid.py                                     | 2 ++
 llava/train/train_llm_to_long.py                                | 2 ++
 llava/train/train_ln.py                                         | 2 ++
 llava/train/train_mem.py                                        | 2 ++
 llava/train/train_mem_ln.py                                     | 2 ++
 llava/train/transformer_normalize_monkey_patch.py               | 2 ++
 llava/train/utils.py                                            | 2 ++
 llava/trl/__init__.py                                           | 2 ++
 llava/trl/core.py                                               | 2 ++
 llava/trl/environment/__init__.py                               | 2 ++
 llava/trl/environment/base_environment.py                       | 2 ++
 llava/trl/extras/__init__.py                                    | 2 ++
 llava/trl/extras/best_of_n_sampler.py                           | 2 ++
 llava/trl/extras/dataset_formatting.py                          | 2 ++
 llava/trl/import_utils.py                                       | 2 ++
 llava/trl/models/__init__.py                                    | 2 ++
 llava/trl/models/modeling_base.py                               | 2 ++
 llava/trl/models/modeling_sd_base.py                            | 2 ++
 llava/trl/models/modeling_value_head.py                         | 2 ++
 llava/trl/models/utils.py                                       | 2 ++
 llava/trl/trainer/__init__.py                                   | 2 ++
 llava/trl/trainer/base.py                                       | 2 ++
 llava/trl/trainer/ddpo_config.py                                | 2 ++
 llava/trl/trainer/ddpo_trainer.py                               | 2 ++
 llava/trl/trainer/dpo_trainer.py                                | 2 ++
 llava/trl/trainer/iterative_sft_trainer.py                      | 2 ++
 llava/trl/trainer/model_config.py                               | 2 ++
 llava/trl/trainer/ppo_config.py                                 | 2 ++
 llava/trl/trainer/ppo_trainer.py                                | 2 ++
 llava/trl/trainer/reward_config.py                              | 2 ++
 llava/trl/trainer/reward_trainer.py                             | 2 ++
 llava/trl/trainer/sft_trainer.py                                | 2 ++
 llava/trl/trainer/utils.py                                      | 2 ++
 llava/utils/__init__.py                                         | 2 ++
 llava/utils/distributed.py                                      | 2 ++
 llava/utils/io.py                                               | 2 ++
 llava/utils/logging.py                                          | 2 ++
 llava/utils/media.py                                            | 2 ++
 llava/utils/merge_lora_weights_and_save_hf_model.py             | 2 ++
 llava/utils/tokenizer.py                                        | 2 ++
 llava/utils/utils.py                                            | 2 ++
 llava/wids/__init__.py                                          | 2 ++
 llava/wids/wids.py                                              | 2 ++
 llava/wids/wids_bench.py                                        | 2 ++
 llava/wids/wids_cleanup.py                                      | 2 ++
 llava/wids/wids_dir.py                                          | 2 ++
 llava/wids/wids_dl.py                                           | 2 ++
 llava/wids/wids_index.py                                        | 2 ++
 llava/wids/wids_lru.py                                          | 2 ++
 llava/wids/wids_mmtar.py                                        | 2 ++
 llava/wids/wids_specs.py                                        | 2 ++
 llava/wids/wids_tar.py                                          | 2 ++
 server.py                                                       | 2 ++
 284 files changed, 568 insertions(+)

diff --git a/data_prepare/coyo/coyo_downloader.py b/data_prepare/coyo/coyo_downloader.py
index 14f46c2c..82192215 100644
--- a/data_prepare/coyo/coyo_downloader.py
+++ b/data_prepare/coyo/coyo_downloader.py
@@ -136,3 +136,5 @@ async def main(data_list):
 
 
 asyncio.run(main(metadata_list))
+
+
diff --git a/data_prepare/coyo/coyo_splitter.py b/data_prepare/coyo/coyo_splitter.py
index 18ac6e14..e53ddc85 100644
--- a/data_prepare/coyo/coyo_splitter.py
+++ b/data_prepare/coyo/coyo_splitter.py
@@ -46,3 +46,5 @@
             f.write(str(len(samples2write)))
 
         counter += 1
+
+
diff --git a/data_prepare/mmc4/mmc4_downloader.py b/data_prepare/mmc4/mmc4_downloader.py
index 17147f39..51ee9f53 100644
--- a/data_prepare/mmc4/mmc4_downloader.py
+++ b/data_prepare/mmc4/mmc4_downloader.py
@@ -151,3 +151,5 @@ async def main(data_list):
 
 
 asyncio.run(main(all_data))
+
+
diff --git a/data_prepare/mmc4/mmc4_filter_and_counter.py b/data_prepare/mmc4/mmc4_filter_and_counter.py
index 6bfd3b82..d1f07060 100644
--- a/data_prepare/mmc4/mmc4_filter_and_counter.py
+++ b/data_prepare/mmc4/mmc4_filter_and_counter.py
@@ -56,3 +56,5 @@
 
     with open(os.path.join(output_path, pkl.replace(".pkl", ".count")), "w") as f:
         f.write(str(len(filtered_annotation)))
+
+
diff --git a/data_prepare/mmc4/mmc4_merger.py b/data_prepare/mmc4/mmc4_merger.py
index d9a0dde3..0ae0bafe 100644
--- a/data_prepare/mmc4/mmc4_merger.py
+++ b/data_prepare/mmc4/mmc4_merger.py
@@ -46,3 +46,5 @@
 
     with open(os.path.join(output_path, shard_name.replace(".jsonl", ".pkl")), "wb") as f:
         pickle.dump(data_list, f)
+
+
diff --git a/data_prepare/panda_split.py b/data_prepare/panda_split.py
index 092df3dc..43d410b2 100644
--- a/data_prepare/panda_split.py
+++ b/data_prepare/panda_split.py
@@ -107,3 +107,5 @@ def split_video_to_clips(
     import fire
 
     fire.Fire(split_video_to_clips)
+
+
diff --git a/data_prepare/sft/ART1_2.py b/data_prepare/sft/ART1_2.py
index d7201bbe..9c8e9fb7 100644
--- a/data_prepare/sft/ART1_2.py
+++ b/data_prepare/sft/ART1_2.py
@@ -86,3 +86,5 @@ def coords_list2bbox(coords_list: List[List[int]], width: int, height: int) -> s
         jsonl_file.write("\n")  # Add a newline after each JSON object
 
 print("Processing complete.")
+
+
diff --git a/data_prepare/sft/ESTVQA.py b/data_prepare/sft/ESTVQA.py
index 3142e445..37adea62 100644
--- a/data_prepare/sft/ESTVQA.py
+++ b/data_prepare/sft/ESTVQA.py
@@ -33,3 +33,5 @@ def is_english(text):
             jsonl_file.write("\n")
 
 print("Processing complete.")
+
+
diff --git a/data_prepare/sft/LSVT.py b/data_prepare/sft/LSVT.py
index bcf1b6b4..14aa0e2e 100644
--- a/data_prepare/sft/LSVT.py
+++ b/data_prepare/sft/LSVT.py
@@ -84,3 +84,5 @@ def coords_list2bbox(coords_list: List[List[int]], width: int, height: int) -> s
             jsonl_file.write("\n")  # Add a newline after each JSON object
 
 print("Processing complete.")
+
+
diff --git a/data_prepare/sft/POIE.py b/data_prepare/sft/POIE.py
index 5657e420..5e243a70 100644
--- a/data_prepare/sft/POIE.py
+++ b/data_prepare/sft/POIE.py
@@ -85,3 +85,5 @@ def coords_list2bbox(coords_list: List[List[int]], width: int, height: int) -> s
                 jsonl_file.write("\n")  # Add a newline after each JSON object
 
 print("Processing complete.")
+
+
diff --git a/data_prepare/sft/ReCTS.py b/data_prepare/sft/ReCTS.py
index 72e3c706..87860b06 100644
--- a/data_prepare/sft/ReCTS.py
+++ b/data_prepare/sft/ReCTS.py
@@ -105,3 +105,5 @@ def coords_list2bbox(coords_list: List[List[int]], width: int, height: int) -> s
             jsonl_file.write("\n")  # Add a newline after each JSON object
 
 print("Processing complete.")
+
+
diff --git a/data_prepare/sft/SROIE.py b/data_prepare/sft/SROIE.py
index a71202c1..663c3a46 100644
--- a/data_prepare/sft/SROIE.py
+++ b/data_prepare/sft/SROIE.py
@@ -66,3 +66,5 @@
 
 # Now 'images' contains all the opened images from the image_root directory
 print(f"Successfully opened {len(images)} out of {len(image_files)} images.")
+
+
diff --git a/data_prepare/sft/merge_idefics2.py b/data_prepare/sft/merge_idefics2.py
index fc07cd29..cea5ce33 100644
--- a/data_prepare/sft/merge_idefics2.py
+++ b/data_prepare/sft/merge_idefics2.py
@@ -47,3 +47,5 @@ def load_jsonl(file_path):
     for item in all_data:
         json.dump(item, f)
         f.write("\n")
+
+
diff --git a/data_prepare/sft/merge_llava_onevision.py b/data_prepare/sft/merge_llava_onevision.py
index ff809d9b..c503b3c9 100644
--- a/data_prepare/sft/merge_llava_onevision.py
+++ b/data_prepare/sft/merge_llava_onevision.py
@@ -99,3 +99,5 @@ def load_jsonl(file_path):
     import fire
 
     fire.Fire(main)
+
+
diff --git a/data_prepare/sft/merge_llava_onevision_eagle.py b/data_prepare/sft/merge_llava_onevision_eagle.py
index 4fa6cea6..35aae1b9 100644
--- a/data_prepare/sft/merge_llava_onevision_eagle.py
+++ b/data_prepare/sft/merge_llava_onevision_eagle.py
@@ -83,3 +83,5 @@ def load_jsonl(file_path):
     import fire
 
     fire.Fire(main)
+
+
diff --git a/data_prepare/sft/mtwi.py b/data_prepare/sft/mtwi.py
index 2851751d..890fc8ee 100644
--- a/data_prepare/sft/mtwi.py
+++ b/data_prepare/sft/mtwi.py
@@ -148,3 +148,5 @@ def clip(x):
         jsonl_file.write("\n")  # Add a newline after each JSON object
 
 print("Processing complete.")
+
+
diff --git a/data_prepare/sft/preprocess_art_shangy.py b/data_prepare/sft/preprocess_art_shangy.py
index 7f0de6c3..c4b6b3ec 100644
--- a/data_prepare/sft/preprocess_art_shangy.py
+++ b/data_prepare/sft/preprocess_art_shangy.py
@@ -87,3 +87,5 @@ def convert_txt_to_jsonl(input_file, output_file):
     output_file = "./art500k_processed.jsonl"
     base_path = "./"
     convert_txt_to_jsonl(input_file, output_file)
+
+
diff --git a/data_prepare/sft/preprocess_cambrian.py b/data_prepare/sft/preprocess_cambrian.py
index 5ebbe344..48ff3bc1 100644
--- a/data_prepare/sft/preprocess_cambrian.py
+++ b/data_prepare/sft/preprocess_cambrian.py
@@ -70,3 +70,5 @@ def check_sample(sample):
 
 with open("cambrian_doc_1275k.json", "w") as f:
     json.dump(cambrian_doc_1275k, f)
+
+
diff --git a/data_prepare/sft/preprocess_cambrian_eagle.py b/data_prepare/sft/preprocess_cambrian_eagle.py
index c979149d..f0444b99 100644
--- a/data_prepare/sft/preprocess_cambrian_eagle.py
+++ b/data_prepare/sft/preprocess_cambrian_eagle.py
@@ -75,3 +75,5 @@ def check_sample(sample):
 
 with open(os.path.join(base_path, "cambrian_adlr_train.json"), "w") as f:
     json.dump(cambrian_eagle, f)
+
+
diff --git a/data_prepare/sft/preprocess_docreason.py b/data_prepare/sft/preprocess_docreason.py
index 8971e64e..eb06fff4 100644
--- a/data_prepare/sft/preprocess_docreason.py
+++ b/data_prepare/sft/preprocess_docreason.py
@@ -29,3 +29,5 @@
 
 with open(json_file_processed, "w") as f:
     json.dump(records, f)
+
+
diff --git a/data_prepare/sft/preprocess_flan.py b/data_prepare/sft/preprocess_flan.py
index efde19b3..0ddb420d 100644
--- a/data_prepare/sft/preprocess_flan.py
+++ b/data_prepare/sft/preprocess_flan.py
@@ -70,3 +70,5 @@
 
 with open(os.path.join(save_path, "text_flan_1m.pkl"), "wb") as f:
     pickle.dump(filtered_samples, f)
+
+
diff --git a/data_prepare/sft/preprocess_idefics2.py b/data_prepare/sft/preprocess_idefics2.py
index 6657c91d..65083279 100644
--- a/data_prepare/sft/preprocess_idefics2.py
+++ b/data_prepare/sft/preprocess_idefics2.py
@@ -130,3 +130,5 @@ def process_dataset(args):
     # Map the process_dataset function to the arguments
     for _ in tqdm(pool.imap_unordered(process_dataset, args), total=len(args), desc="Processing datasets"):
         pass
+
+
diff --git a/data_prepare/sft/preprocess_idefics2_eagle.py b/data_prepare/sft/preprocess_idefics2_eagle.py
index 175d30cb..8354d35b 100644
--- a/data_prepare/sft/preprocess_idefics2_eagle.py
+++ b/data_prepare/sft/preprocess_idefics2_eagle.py
@@ -170,3 +170,5 @@ def main(
     import fire
 
     fire.Fire(main)
+
+
diff --git a/data_prepare/sft/preprocess_kvqa.py b/data_prepare/sft/preprocess_kvqa.py
index 92c07244..e024882f 100644
--- a/data_prepare/sft/preprocess_kvqa.py
+++ b/data_prepare/sft/preprocess_kvqa.py
@@ -45,3 +45,5 @@
     json.dump(new_records, f)
 
 print(len(new_records))
+
+
diff --git a/data_prepare/sft/preprocess_llava_onevision.py b/data_prepare/sft/preprocess_llava_onevision.py
index 8ca52fb7..1f741861 100644
--- a/data_prepare/sft/preprocess_llava_onevision.py
+++ b/data_prepare/sft/preprocess_llava_onevision.py
@@ -103,3 +103,5 @@ def main(
     import fire
 
     fire.Fire(main)
+
+
diff --git a/data_prepare/sft/preprocess_m3it.py b/data_prepare/sft/preprocess_m3it.py
index 19b16a2f..b7d954b2 100644
--- a/data_prepare/sft/preprocess_m3it.py
+++ b/data_prepare/sft/preprocess_m3it.py
@@ -82,3 +82,5 @@
     save_filename = os.path.join(save_path, save_filename)
     with open(save_filename, "wb") as f:
         pickle.dump(dataset, f)
+
+
diff --git a/data_prepare/sft/preprocess_metamathqa.py b/data_prepare/sft/preprocess_metamathqa.py
index e3ee52e9..a5151db4 100644
--- a/data_prepare/sft/preprocess_metamathqa.py
+++ b/data_prepare/sft/preprocess_metamathqa.py
@@ -25,3 +25,5 @@
 
 with open(json_file_processed, "w") as f:
     json.dump(records, f)
+
+
diff --git a/data_prepare/sft/preprocess_viquae.py b/data_prepare/sft/preprocess_viquae.py
index 3487dac7..6d933c50 100644
--- a/data_prepare/sft/preprocess_viquae.py
+++ b/data_prepare/sft/preprocess_viquae.py
@@ -45,3 +45,5 @@ def base64_to_pil_image(base64_string):
 
 with open(os.path.join(base_path, "viquae_processed.json"), "w") as f:
     json.dump(new_records, f)
+
+
diff --git a/data_prepare/sft/split_vflan.py b/data_prepare/sft/split_vflan.py
index 93d0715f..13a78053 100644
--- a/data_prepare/sft/split_vflan.py
+++ b/data_prepare/sft/split_vflan.py
@@ -57,3 +57,5 @@
     print(f"Finished writing part-{counter:05d}.pkl!")
 
     counter += 1
+
+
diff --git a/data_prepare/sft/unichart_pretrain.py b/data_prepare/sft/unichart_pretrain.py
index a544a829..47b8b3cd 100644
--- a/data_prepare/sft/unichart_pretrain.py
+++ b/data_prepare/sft/unichart_pretrain.py
@@ -57,3 +57,5 @@
                 pbar.update(1)
 
 print("Processing complete.")
+
+
diff --git a/data_prepare/sft/unichart_sft.py b/data_prepare/sft/unichart_sft.py
index 81e46c70..3bb29054 100644
--- a/data_prepare/sft/unichart_sft.py
+++ b/data_prepare/sft/unichart_sft.py
@@ -59,3 +59,5 @@
                 pbar.update(1)
 
 print("Processing complete.")
+
+
diff --git a/llava/__init__.py b/llava/__init__.py
index 37a40560..1d61b293 100755
--- a/llava/__init__.py
+++ b/llava/__init__.py
@@ -1,2 +1,4 @@
 from .entry import *
 from .media import *
+
+
diff --git a/llava/cli/eval.py b/llava/cli/eval.py
index 53939abb..78997e46 100644
--- a/llava/cli/eval.py
+++ b/llava/cli/eval.py
@@ -198,3 +198,5 @@ def main() -> None:
 
 if __name__ == "__main__":
     main()
+
+
diff --git a/llava/cli/infer.py b/llava/cli/infer.py
index b45ecbdd..f4c74c01 100644
--- a/llava/cli/infer.py
+++ b/llava/cli/infer.py
@@ -42,3 +42,5 @@ def main() -> None:
 
 if __name__ == "__main__":
     main()
+
+
diff --git a/llava/cli/run.py b/llava/cli/run.py
index e2fea74a..8af114f5 100644
--- a/llava/cli/run.py
+++ b/llava/cli/run.py
@@ -132,3 +132,5 @@ def main() -> None:
 
 if __name__ == "__main__":
     main()
+
+
diff --git a/llava/cli/upload2hf.py b/llava/cli/upload2hf.py
index 5387915b..ca6d0e62 100644
--- a/llava/cli/upload2hf.py
+++ b/llava/cli/upload2hf.py
@@ -222,3 +222,5 @@ def main():
 
 if __name__ == "__main__":
     main()
+
+
diff --git a/llava/constants.py b/llava/constants.py
index bb1ff810..84385fdf 100755
--- a/llava/constants.py
+++ b/llava/constants.py
@@ -30,3 +30,5 @@
     "image": "<image>",
     "video": "<vila/video>",
 }
+
+
diff --git a/llava/conversation.py b/llava/conversation.py
index 0d8b857a..e9572392 100755
--- a/llava/conversation.py
+++ b/llava/conversation.py
@@ -189,3 +189,5 @@ def auto_set_conversation_mode(model_name_or_path: str) -> str:
             logger.info(f"Setting conversation mode to `{v}` based on model name/path `{model_name_or_path}`.")
             default_conversation = conv_templates[v]
             return
+
+
diff --git a/llava/data/__init__.py b/llava/data/__init__.py
index 87b1dc8b..d4eadccb 100755
--- a/llava/data/__init__.py
+++ b/llava/data/__init__.py
@@ -3,3 +3,5 @@
 from .dataset_impl import *
 from .datasets_mixture import *
 from .simple_vila_webdataset import VILAWebDataset
+
+
diff --git a/llava/data/base.py b/llava/data/base.py
index a4cbcf19..8450b744 100644
--- a/llava/data/base.py
+++ b/llava/data/base.py
@@ -84,3 +84,5 @@ def __getitem__(self, index: int) -> Dict[str, Any]:
 
     def __len__(self) -> int:
         return len(self.instances)
+
+
diff --git a/llava/data/builder.py b/llava/data/builder.py
index 54c7a05f..ae1ae967 100644
--- a/llava/data/builder.py
+++ b/llava/data/builder.py
@@ -213,3 +213,5 @@ def build_dataset_legacy(
         data_args=data_args,
         training_args=training_args,
     )
+
+
diff --git a/llava/data/collate.py b/llava/data/collate.py
index fac3fe5f..193f5afd 100644
--- a/llava/data/collate.py
+++ b/llava/data/collate.py
@@ -88,3 +88,5 @@ def __call__(self, instances: Sequence[Dict[str, Any]]) -> Dict[str, Any]:
             "labels": labels,
             "attention_mask": attention_mask,
         }
+
+
diff --git a/llava/data/dataset.py b/llava/data/dataset.py
index 54679038..2659d735 100755
--- a/llava/data/dataset.py
+++ b/llava/data/dataset.py
@@ -1576,3 +1576,5 @@ def make_supervised_data_module(
         train_dataset=train_dataset,
         data_collator=data_collator,
     )
+
+
diff --git a/llava/data/dataset_impl/__init__.py b/llava/data/dataset_impl/__init__.py
index d5a8bc7f..107173e2 100644
--- a/llava/data/dataset_impl/__init__.py
+++ b/llava/data/dataset_impl/__init__.py
@@ -2,3 +2,5 @@
 from .lita import *
 from .llava import *
 from .llava_cot import *
+
+
diff --git a/llava/data/dataset_impl/coyo_qa.py b/llava/data/dataset_impl/coyo_qa.py
index 18666457..ce764fcb 100644
--- a/llava/data/dataset_impl/coyo_qa.py
+++ b/llava/data/dataset_impl/coyo_qa.py
@@ -203,3 +203,5 @@ def __getitem__(self, i) -> Dict[str, torch.Tensor]:
             data_dict["block_sizes"] = block_sizes
 
         return data_dict
+
+
diff --git a/llava/data/dataset_impl/coyo_recap.py b/llava/data/dataset_impl/coyo_recap.py
index caff3a5c..8979799c 100755
--- a/llava/data/dataset_impl/coyo_recap.py
+++ b/llava/data/dataset_impl/coyo_recap.py
@@ -88,3 +88,5 @@ def __init__(
         else:
             self.caption_choice = data_args.caption_choice
         print(f"Current caption choice: {self.caption_choice}.")
+
+
diff --git a/llava/data/dataset_impl/dummy.py b/llava/data/dataset_impl/dummy.py
index 007d569f..2173f0cc 100644
--- a/llava/data/dataset_impl/dummy.py
+++ b/llava/data/dataset_impl/dummy.py
@@ -92,3 +92,5 @@ def process(self, instance: Dict[str, Any]) -> List[Dict[str, Any]]:
         # Add media to the beginning of the first message
         messages[0]["value"] = medias + [messages[0]["value"]]
         return messages
+
+
diff --git a/llava/data/dataset_impl/lita.py b/llava/data/dataset_impl/lita.py
index 9788b397..561a55bc 100644
--- a/llava/data/dataset_impl/lita.py
+++ b/llava/data/dataset_impl/lita.py
@@ -241,3 +241,5 @@ def process(self, instance: Dict[str, Any]) -> List[Dict[str, Any]]:
         video = Video(instance["video_path"])
         messages[0]["value"] = [video, messages[0]["value"]]
         return messages
+
+
diff --git a/llava/data/dataset_impl/llava.py b/llava/data/dataset_impl/llava.py
index c7041fa6..78b5c0c6 100644
--- a/llava/data/dataset_impl/llava.py
+++ b/llava/data/dataset_impl/llava.py
@@ -132,3 +132,5 @@ def process(self, instance: Dict[str, Any]) -> List[Dict[str, Any]]:
             new_value = [*img_list, value.replace(DEFAULT_IMAGE_TOKEN, "").strip()]
             messages[0]["value"] = new_value
         return messages
+
+
diff --git a/llava/data/dataset_impl/llava_cot.py b/llava/data/dataset_impl/llava_cot.py
index 61f7bd1d..fbf52c10 100644
--- a/llava/data/dataset_impl/llava_cot.py
+++ b/llava/data/dataset_impl/llava_cot.py
@@ -174,3 +174,5 @@ def process_multi_img(self, instance: Dict[str, Any], index: int) -> List[Dict[s
     assert len(medias) == 0, f"#Num of <images> does not match the number of images in the instance. {instance}"
 
     return messages
+
+
diff --git a/llava/data/dataset_impl/panda70m.py b/llava/data/dataset_impl/panda70m.py
index 3d9a2dae..47123e88 100755
--- a/llava/data/dataset_impl/panda70m.py
+++ b/llava/data/dataset_impl/panda70m.py
@@ -229,3 +229,5 @@ def cleanup_corrupted_videos(
     jinfo = json.load(open(json_path))
     img_t = load_video(video_path, jinfo=jinfo)
     print(img_t)
+
+
diff --git a/llava/data/dataset_impl/sam.py b/llava/data/dataset_impl/sam.py
index 604d8067..cc68597b 100755
--- a/llava/data/dataset_impl/sam.py
+++ b/llava/data/dataset_impl/sam.py
@@ -230,3 +230,5 @@ def __getitem__(self, i) -> Dict[str, torch.Tensor]:
     for idx, data in enumerate(dst):
         print(idx, data.keys())
 # nvcode: off
+
+
diff --git a/llava/data/dataset_impl/textocr.py b/llava/data/dataset_impl/textocr.py
index aa37ba32..40c38b94 100755
--- a/llava/data/dataset_impl/textocr.py
+++ b/llava/data/dataset_impl/textocr.py
@@ -289,3 +289,5 @@ def __getitem__(self, index):
 
     for idx in range(2):
         pprint(dataset[idx])
+
+
diff --git a/llava/data/dataset_impl/utils.py b/llava/data/dataset_impl/utils.py
index e3dc986f..8870ffd4 100644
--- a/llava/data/dataset_impl/utils.py
+++ b/llava/data/dataset_impl/utils.py
@@ -27,3 +27,5 @@ def _remove_media_tokens(text: str) -> str:
     for token in ["<image>", "<video>"]:
         text = text.replace(token + "\n", "").replace("\n" + token, "").replace(token, "")
     return text.strip()
+
+
diff --git a/llava/data/datasets_mixture.py b/llava/data/datasets_mixture.py
index 9e0d17a1..6ee5e7c2 100755
--- a/llava/data/datasets_mixture.py
+++ b/llava/data/datasets_mixture.py
@@ -54,3 +54,5 @@ def add_dataset(dataset):
 
 def register_datasets_mixtures():
     pass
+
+
diff --git a/llava/data/simple_vila_webdataset.py b/llava/data/simple_vila_webdataset.py
index 675715f2..637c19d8 100755
--- a/llava/data/simple_vila_webdataset.py
+++ b/llava/data/simple_vila_webdataset.py
@@ -334,3 +334,5 @@ def merge(a: dict, b: dict, path=[], strict=False):
 
             # if idx >= 5:
             #     break
+
+
diff --git a/llava/data/utils.py b/llava/data/utils.py
index 9572f530..f9e49d03 100755
--- a/llava/data/utils.py
+++ b/llava/data/utils.py
@@ -35,3 +35,5 @@ def from_bytesio(cls, file_path: str, decode_audio: bool = True, decoder: str =
             print(f"unsupported type {type(file_path)}")
         video_cls = select_video_class(decoder)
         return video_cls(video_file, pathlib.Path(file_path).name, decode_audio)
+
+
diff --git a/llava/entry.py b/llava/entry.py
index bb2f105c..809e375d 100644
--- a/llava/entry.py
+++ b/llava/entry.py
@@ -36,3 +36,5 @@ def load(
 
     model = load_pretrained_model(model_path, model_name, model_base, **kwargs)[1]
     return model
+
+
diff --git a/llava/eval/__init__.py b/llava/eval/__init__.py
index 76d22cb1..fc37c509 100644
--- a/llava/eval/__init__.py
+++ b/llava/eval/__init__.py
@@ -7,3 +7,5 @@
 
 EVAL_ROOT = "scripts/eval"
 TASKS = io.load(os.path.join(os.path.dirname(__file__), "registry.yaml"))
+
+
diff --git a/llava/eval/cinepile.py b/llava/eval/cinepile.py
index 45215873..084996fb 100644
--- a/llava/eval/cinepile.py
+++ b/llava/eval/cinepile.py
@@ -106,3 +106,5 @@ def main() -> None:
 
 if __name__ == "__main__":
     main()
+
+
diff --git a/llava/eval/egoschema.py b/llava/eval/egoschema.py
index 403be090..f5e03c43 100644
--- a/llava/eval/egoschema.py
+++ b/llava/eval/egoschema.py
@@ -107,3 +107,5 @@ def main() -> None:
 
 if __name__ == "__main__":
     main()
+
+
diff --git a/llava/eval/eval_refcoco.py b/llava/eval/eval_refcoco.py
index 9ea35243..ee65a837 100755
--- a/llava/eval/eval_refcoco.py
+++ b/llava/eval/eval_refcoco.py
@@ -71,3 +71,5 @@ def computeIoU(bbox1, bbox2):
             except Exception as e:
                 print(e, flush=True)
                 continue
+
+
diff --git a/llava/eval/eventbench.py b/llava/eval/eventbench.py
index 258a8179..0da9b485 100644
--- a/llava/eval/eventbench.py
+++ b/llava/eval/eventbench.py
@@ -91,3 +91,5 @@ def main() -> None:
 
 if __name__ == "__main__":
     main()
+
+
diff --git a/llava/eval/lmms/models/__init__.py b/llava/eval/lmms/models/__init__.py
index a965cc88..b6adb217 100644
--- a/llava/eval/lmms/models/__init__.py
+++ b/llava/eval/lmms/models/__init__.py
@@ -1,3 +1,5 @@
 AVAILABLE_MODELS = {
     "vila_internal": "VILA",
 }
+
+
diff --git a/llava/eval/lmms/models/vila_internal.py b/llava/eval/lmms/models/vila_internal.py
index 973a6e7f..77e045f0 100644
--- a/llava/eval/lmms/models/vila_internal.py
+++ b/llava/eval/lmms/models/vila_internal.py
@@ -139,3 +139,5 @@ def generate_until_multi_round(self, requests: List[Instance]) -> List[str]:
 
     def loglikelihood(self, requests: List[Instance]) -> List[Tuple[float, bool]]:
         raise NotImplementedError
+
+
diff --git a/llava/eval/lmms/tasks/__init__.py b/llava/eval/lmms/tasks/__init__.py
index e69de29b..139597f9 100644
--- a/llava/eval/lmms/tasks/__init__.py
+++ b/llava/eval/lmms/tasks/__init__.py
@@ -0,0 +1,2 @@
+
+
diff --git a/llava/eval/lmms/tasks/videomme.py b/llava/eval/lmms/tasks/videomme.py
index 1e689b1a..c1dfaf66 100644
--- a/llava/eval/lmms/tasks/videomme.py
+++ b/llava/eval/lmms/tasks/videomme.py
@@ -64,3 +64,5 @@ def videomme_doc_to_text_subtitle(doc: Dict[str, Any], num_frames: int) -> str:
     prompt += "\n".join(doc["options"]) + "\n"
     prompt += "The best answer is:"
     return prompt
+
+
diff --git a/llava/eval/m4c_evaluator.py b/llava/eval/m4c_evaluator.py
index 9f6225a8..2b25c91a 100755
--- a/llava/eval/m4c_evaluator.py
+++ b/llava/eval/m4c_evaluator.py
@@ -326,3 +326,5 @@ def eval_pred_list(self, pred_list):
 
         bleu4 = score[3]  # score is (Bleu-1, Bleu-2, Bleu-3, Bleu-4)
         return bleu4
+
+
diff --git a/llava/eval/mathvista.py b/llava/eval/mathvista.py
index 0038b13e..1528acd5 100644
--- a/llava/eval/mathvista.py
+++ b/llava/eval/mathvista.py
@@ -85,3 +85,5 @@ def main():
 
 if __name__ == "__main__":
     main()
+
+
diff --git a/llava/eval/mathvista_utils/calculate_score.py b/llava/eval/mathvista_utils/calculate_score.py
index b424e804..2c262aa2 100755
--- a/llava/eval/mathvista_utils/calculate_score.py
+++ b/llava/eval/mathvista_utils/calculate_score.py
@@ -194,3 +194,5 @@ def simple_calculate_score(results: dict):
             )
         )
     return scores
+
+
diff --git a/llava/eval/mathvista_utils/extract_answer.py b/llava/eval/mathvista_utils/extract_answer.py
index 39e6528f..09bfad98 100755
--- a/llava/eval/mathvista_utils/extract_answer.py
+++ b/llava/eval/mathvista_utils/extract_answer.py
@@ -79,3 +79,5 @@ def extract_answer(response, problem, quick_extract=False):
         print(f"Error in extracting answer for {pid}")
 
     return ""
+
+
diff --git a/llava/eval/mathvista_utils/prompts/ext_ans.py b/llava/eval/mathvista_utils/prompts/ext_ans.py
index 0e996328..a83c38df 100755
--- a/llava/eval/mathvista_utils/prompts/ext_ans.py
+++ b/llava/eval/mathvista_utils/prompts/ext_ans.py
@@ -38,3 +38,5 @@
 
 Extracted answer: B
 """
+
+
diff --git a/llava/eval/mathvista_utils/utilities.py b/llava/eval/mathvista_utils/utilities.py
index b00f0445..5ddd20b2 100755
--- a/llava/eval/mathvista_utils/utilities.py
+++ b/llava/eval/mathvista_utils/utilities.py
@@ -253,3 +253,5 @@ def get_chat_response(
             if sleep_time > 0:
                 time.sleep(sleep_time)
     return ""
+
+
diff --git a/llava/eval/mmmu_utils/eval_utils.py b/llava/eval/mmmu_utils/eval_utils.py
index 42c3cc6b..ee84e96f 100755
--- a/llava/eval/mmmu_utils/eval_utils.py
+++ b/llava/eval/mmmu_utils/eval_utils.py
@@ -59,3 +59,5 @@ def parse_choice(response, all_choices, index2ans=None):
         pred_index = candidates[0]
 
     return pred_index
+
+
diff --git a/llava/eval/model_refcoco.py b/llava/eval/model_refcoco.py
index c5e0465a..991b3193 100755
--- a/llava/eval/model_refcoco.py
+++ b/llava/eval/model_refcoco.py
@@ -377,3 +377,5 @@ def eval_model(args):
     args = parser.parse_args()
 
     eval_model(args)
+
+
diff --git a/llava/eval/model_vqa_video.py b/llava/eval/model_vqa_video.py
index 8940422b..6c7c67fb 100755
--- a/llava/eval/model_vqa_video.py
+++ b/llava/eval/model_vqa_video.py
@@ -209,3 +209,5 @@ def eval_model(args):
     args = parser.parse_args()
 
     eval_model(args)
+
+
diff --git a/llava/eval/model_vqa_videodemo.py b/llava/eval/model_vqa_videodemo.py
index 5c4fef8e..cb39765f 100755
--- a/llava/eval/model_vqa_videodemo.py
+++ b/llava/eval/model_vqa_videodemo.py
@@ -216,3 +216,5 @@ def eval_model(args):
     args = parser.parse_args()
 
     eval_model(args)
+
+
diff --git a/llava/eval/rtl.py b/llava/eval/rtl.py
index 36f393db..e13ab05f 100644
--- a/llava/eval/rtl.py
+++ b/llava/eval/rtl.py
@@ -133,3 +133,5 @@ def main() -> None:
 
 if __name__ == "__main__":
     main()
+
+
diff --git a/llava/eval/textvqa.py b/llava/eval/textvqa.py
index 5f9e17f9..9e6452cc 100755
--- a/llava/eval/textvqa.py
+++ b/llava/eval/textvqa.py
@@ -120,3 +120,5 @@ def main() -> None:
 
 if __name__ == "__main__":
     main()
+
+
diff --git a/llava/eval/video/eval_benchmark_1_correctness.py b/llava/eval/video/eval_benchmark_1_correctness.py
index fb3779d7..c7a16638 100644
--- a/llava/eval/video/eval_benchmark_1_correctness.py
+++ b/llava/eval/video/eval_benchmark_1_correctness.py
@@ -215,3 +215,5 @@ def main():
     main()
     end_time = time.time()
     print(f"took {end_time - start_time} seconds")
+
+
diff --git a/llava/eval/video/eval_benchmark_2_detailed_orientation.py b/llava/eval/video/eval_benchmark_2_detailed_orientation.py
index 8eec5113..c63a90f4 100644
--- a/llava/eval/video/eval_benchmark_2_detailed_orientation.py
+++ b/llava/eval/video/eval_benchmark_2_detailed_orientation.py
@@ -207,3 +207,5 @@ def main():
 
 if __name__ == "__main__":
     main()
+
+
diff --git a/llava/eval/video/eval_benchmark_3_context.py b/llava/eval/video/eval_benchmark_3_context.py
index d0a5bfc1..36fb86aa 100644
--- a/llava/eval/video/eval_benchmark_3_context.py
+++ b/llava/eval/video/eval_benchmark_3_context.py
@@ -208,3 +208,5 @@ def main():
 
 if __name__ == "__main__":
     main()
+
+
diff --git a/llava/eval/video/eval_benchmark_4_temporal.py b/llava/eval/video/eval_benchmark_4_temporal.py
index 059d157c..4b5f736b 100644
--- a/llava/eval/video/eval_benchmark_4_temporal.py
+++ b/llava/eval/video/eval_benchmark_4_temporal.py
@@ -206,3 +206,5 @@ def main():
 
 if __name__ == "__main__":
     main()
+
+
diff --git a/llava/eval/video/eval_benchmark_5_consistency.py b/llava/eval/video/eval_benchmark_5_consistency.py
index 4bf9a829..c681434e 100644
--- a/llava/eval/video/eval_benchmark_5_consistency.py
+++ b/llava/eval/video/eval_benchmark_5_consistency.py
@@ -214,3 +214,5 @@ def main():
 
 if __name__ == "__main__":
     main()
+
+
diff --git a/llava/eval/video/eval_video_qa.py b/llava/eval/video/eval_video_qa.py
index 07950560..7b961ee5 100755
--- a/llava/eval/video/eval_video_qa.py
+++ b/llava/eval/video/eval_video_qa.py
@@ -223,3 +223,5 @@ def main():
 
 if __name__ == "__main__":
     main()
+
+
diff --git a/llava/eval/video/model_vqa_videodemo_benchmark.py b/llava/eval/video/model_vqa_videodemo_benchmark.py
index baa1102c..169ab4bb 100644
--- a/llava/eval/video/model_vqa_videodemo_benchmark.py
+++ b/llava/eval/video/model_vqa_videodemo_benchmark.py
@@ -212,3 +212,5 @@ def gen_pred_json(output_dir, eval_type):
     args = parser.parse_args()
 
     eval_model(args)
+
+
diff --git a/llava/eval/video/utils.py b/llava/eval/video/utils.py
index 96a5785d..b6035fd4 100644
--- a/llava/eval/video/utils.py
+++ b/llava/eval/video/utils.py
@@ -21,3 +21,5 @@ def get_client():
         raise ValueError("At least provide one format for gpt assisted benchmarking")
 
     return client
+
+
diff --git a/llava/eval/vision_niah_vila/eval_vision_niah.py b/llava/eval/vision_niah_vila/eval_vision_niah.py
index d3fd6c4e..ba37d599 100644
--- a/llava/eval/vision_niah_vila/eval_vision_niah.py
+++ b/llava/eval/vision_niah_vila/eval_vision_niah.py
@@ -448,3 +448,5 @@ def main(args):
     args.add_argument("--plot_only", action="store_true")
 
     main(args.parse_args())
+
+
diff --git a/llava/eval/vision_niah_vila/produce_haystack_embedding.py b/llava/eval/vision_niah_vila/produce_haystack_embedding.py
index db26daa5..b275dc16 100644
--- a/llava/eval/vision_niah_vila/produce_haystack_embedding.py
+++ b/llava/eval/vision_niah_vila/produce_haystack_embedding.py
@@ -104,3 +104,5 @@ def main(args):
     parser.add_argument("--add_newline_token", action="store_true")
     args = parser.parse_args()
     main(args)
+
+
diff --git a/llava/eval/vision_niah_vila/produce_needle_embedding.py b/llava/eval/vision_niah_vila/produce_needle_embedding.py
index 7e5963db..3aa2b878 100644
--- a/llava/eval/vision_niah_vila/produce_needle_embedding.py
+++ b/llava/eval/vision_niah_vila/produce_needle_embedding.py
@@ -68,3 +68,5 @@ def main(args):
     parser.add_argument("--pooling_size", type=int, default=0)
     args = parser.parse_args()
     main(args)
+
+
diff --git a/llava/eval/vision_niah_vila/zigzag_ring_attn/modeling_qwen2.py b/llava/eval/vision_niah_vila/zigzag_ring_attn/modeling_qwen2.py
index d0f446cb..1359f615 100644
--- a/llava/eval/vision_niah_vila/zigzag_ring_attn/modeling_qwen2.py
+++ b/llava/eval/vision_niah_vila/zigzag_ring_attn/modeling_qwen2.py
@@ -1395,3 +1395,5 @@ def forward(
             hidden_states=outputs.hidden_states,
             attentions=outputs.attentions,
         )
+
+
diff --git a/llava/eval/vision_niah_vila/zigzag_ring_attn/monkey_patch.py b/llava/eval/vision_niah_vila/zigzag_ring_attn/monkey_patch.py
index 57dea659..8acb6ef3 100644
--- a/llava/eval/vision_niah_vila/zigzag_ring_attn/monkey_patch.py
+++ b/llava/eval/vision_niah_vila/zigzag_ring_attn/monkey_patch.py
@@ -123,3 +123,5 @@ def apply_zigzag_ring_attn_monkey_patch_mistral():
         new_flash_attn_forward
     )
     transformers.models.mistral.modeling_mistral.MistralDecoderLayer.forward = new_decoder_forward
+
+
diff --git a/llava/eval/vision_niah_vila/zigzag_ring_attn/prepare_inputs.py b/llava/eval/vision_niah_vila/zigzag_ring_attn/prepare_inputs.py
index 54018edf..ec57aba9 100644
--- a/llava/eval/vision_niah_vila/zigzag_ring_attn/prepare_inputs.py
+++ b/llava/eval/vision_niah_vila/zigzag_ring_attn/prepare_inputs.py
@@ -51,3 +51,5 @@ def prepare_zigzag_ring_attn_inputs(input_ids, position_ids, target_ids, rank, w
         "local_position_ids": local_position_ids,
         "local_target_ids": local_target_ids,
     }
+
+
diff --git a/llava/eval/vnbench.py b/llava/eval/vnbench.py
index c5f898fe..15cba647 100644
--- a/llava/eval/vnbench.py
+++ b/llava/eval/vnbench.py
@@ -113,3 +113,5 @@ def main() -> None:
 
 if __name__ == "__main__":
     main()
+
+
diff --git a/llava/media.py b/llava/media.py
index 8c82b17f..cc4327c9 100644
--- a/llava/media.py
+++ b/llava/media.py
@@ -16,3 +16,5 @@ class Image(File):
 
 class Video(File):
     pass
+
+
diff --git a/llava/mm_utils.py b/llava/mm_utils.py
index cd29f5d7..c7b0f1ec 100755
--- a/llava/mm_utils.py
+++ b/llava/mm_utils.py
@@ -567,3 +567,5 @@ def __call__(self, output_ids: torch.LongTensor, scores: torch.FloatTensor, **kw
         for i in range(output_ids.shape[0]):
             outputs.append(self.call_for_batch(output_ids[i].unsqueeze(0), scores))
         return all(outputs)
+
+
diff --git a/llava/model/FloatPointQuantizeTorch.py b/llava/model/FloatPointQuantizeTorch.py
index 1775b9b6..674410c8 100644
--- a/llava/model/FloatPointQuantizeTorch.py
+++ b/llava/model/FloatPointQuantizeTorch.py
@@ -77,3 +77,5 @@ def ZeroDynamic_quantize_torch(x, bit, stochastic):
 
     y = y.to(x)
     return y
+
+
diff --git a/llava/model/FloatPointQuantizeTriton.py b/llava/model/FloatPointQuantizeTriton.py
index cd07b706..855e80e5 100644
--- a/llava/model/FloatPointQuantizeTriton.py
+++ b/llava/model/FloatPointQuantizeTriton.py
@@ -191,3 +191,5 @@ def _floatExMy_stochastic_quantize_kernel(
     y = y.to(x_ptr.dtype.element_ty)
 
     tl.store(output_ptr + offsets, y, mask=mask)
+
+
diff --git a/llava/model/__init__.py b/llava/model/__init__.py
index 6b69e366..26d16610 100755
--- a/llava/model/__init__.py
+++ b/llava/model/__init__.py
@@ -27,3 +27,5 @@
 """
 from .language_model.fp8linearqwen2 import FP8LinearQwen2Config, FP8LinearQwen2Model
 from .language_model.qllava_qllama import QLlavaLlamaConfig, QLlavaLlamaModel
+
+
diff --git a/llava/model/apply_delta.py b/llava/model/apply_delta.py
index 1b547539..74a99255 100755
--- a/llava/model/apply_delta.py
+++ b/llava/model/apply_delta.py
@@ -69,3 +69,5 @@ def apply_delta(base_model_path, target_model_path, delta_path):
     args = parser.parse_args()
 
     apply_delta(args.base_model_path, args.target_model_path, args.delta_path)
+
+
diff --git a/llava/model/builder.py b/llava/model/builder.py
index ddff6835..df10e2a5 100755
--- a/llava/model/builder.py
+++ b/llava/model/builder.py
@@ -160,3 +160,5 @@ def prepare_config_for_eval(config: PretrainedConfig, kwargs: dict):
         raise ValueError(f"Invalid configuration! Cannot find vision_tower in config:\n{config}")
 
     config.model_dtype = kwargs.pop("torch_dtype").__str__()
+
+
diff --git a/llava/model/coat/activation/__init__.py b/llava/model/coat/activation/__init__.py
index e69de29b..139597f9 100644
--- a/llava/model/coat/activation/__init__.py
+++ b/llava/model/coat/activation/__init__.py
@@ -0,0 +1,2 @@
+
+
diff --git a/llava/model/coat/activation/fake_quantization/FloatPointQuantizeTorch.py b/llava/model/coat/activation/fake_quantization/FloatPointQuantizeTorch.py
index e5b20c2f..0e90adef 100644
--- a/llava/model/coat/activation/fake_quantization/FloatPointQuantizeTorch.py
+++ b/llava/model/coat/activation/fake_quantization/FloatPointQuantizeTorch.py
@@ -93,3 +93,5 @@ def ZeroDynamic_quantize_torch(x, bit, stochastic):
 
     y = y.to(x)
     return y
+
+
diff --git a/llava/model/coat/activation/fake_quantization/FloatPointQuantizeTriton.py b/llava/model/coat/activation/fake_quantization/FloatPointQuantizeTriton.py
index 6560ba8a..27ffa1fa 100644
--- a/llava/model/coat/activation/fake_quantization/FloatPointQuantizeTriton.py
+++ b/llava/model/coat/activation/fake_quantization/FloatPointQuantizeTriton.py
@@ -173,3 +173,5 @@ def _floatExMy_stochastic_quantize_kernel(
     y = y.to(x_ptr.dtype.element_ty)
 
     tl.store(output_ptr + offsets, y, mask=mask)
+
+
diff --git a/llava/model/coat/activation/fake_quantization/quantize_function.py b/llava/model/coat/activation/fake_quantization/quantize_function.py
index 4db820ab..ae60c53b 100644
--- a/llava/model/coat/activation/fake_quantization/quantize_function.py
+++ b/llava/model/coat/activation/fake_quantization/quantize_function.py
@@ -231,3 +231,5 @@ def forward(ctx, input, symm, bits, stochastic, epsilon, apply_quantize=True, la
     @staticmethod
     def backward(ctx, grad_output):
         return grad_output, None, None, None, None, None
+
+
diff --git a/llava/model/coat/activation/fake_quantization/utils.py b/llava/model/coat/activation/fake_quantization/utils.py
index a8034d8d..b7466808 100644
--- a/llava/model/coat/activation/fake_quantization/utils.py
+++ b/llava/model/coat/activation/fake_quantization/utils.py
@@ -107,3 +107,5 @@ def flatten_to_1d(tensor):
         return None
 
     return tensor.reshape(-1)
+
+
diff --git a/llava/model/coat/activation/models/_fp8_quantization_config.py b/llava/model/coat/activation/models/_fp8_quantization_config.py
index c2a8c894..96f527e4 100644
--- a/llava/model/coat/activation/models/_fp8_quantization_config.py
+++ b/llava/model/coat/activation/models/_fp8_quantization_config.py
@@ -59,3 +59,5 @@ def __init__(
 
         self.row_blocksize = row_blocksize
         self.col_blocksize = col_blocksize
+
+
diff --git a/llava/model/coat/activation/models/_fp8_weightcache.py b/llava/model/coat/activation/models/_fp8_weightcache.py
index aa7076da..7f1ff84a 100644
--- a/llava/model/coat/activation/models/_fp8_weightcache.py
+++ b/llava/model/coat/activation/models/_fp8_weightcache.py
@@ -40,3 +40,5 @@ def prepare_weight(self, weight, weight_name, is_first_microbatch):
 
     def forward(self, x):
         pass
+
+
diff --git a/llava/model/coat/activation/models/_fp8manager.py b/llava/model/coat/activation/models/_fp8manager.py
index 2d0a84c7..81db9aa6 100644
--- a/llava/model/coat/activation/models/_fp8manager.py
+++ b/llava/model/coat/activation/models/_fp8manager.py
@@ -23,3 +23,5 @@ class FP8Manager:
     """
 
     is_first_microbatch = False
+
+
diff --git a/llava/model/coat/activation/models/coat_llama.py b/llava/model/coat/activation/models/coat_llama.py
index 4773e35d..6edfc6b4 100644
--- a/llava/model/coat/activation/models/coat_llama.py
+++ b/llava/model/coat/activation/models/coat_llama.py
@@ -1471,3 +1471,5 @@ def make_state_dict_compatible(state_dict: dict[str, torch.Tensor]):
 AutoConfig.register("fp8_llama", CoatLlamaConfig)
 AutoModel.register(CoatLlamaConfig, CoatLlamaModel)
 AutoModelForCausalLM.register(CoatLlamaConfig, CoatLlamaForCausalLM)
+
+
diff --git a/llava/model/coat/activation/models/coat_llama_convert_from_hf.py b/llava/model/coat/activation/models/coat_llama_convert_from_hf.py
index 465261e7..b2abef15 100644
--- a/llava/model/coat/activation/models/coat_llama_convert_from_hf.py
+++ b/llava/model/coat/activation/models/coat_llama_convert_from_hf.py
@@ -63,3 +63,5 @@ def download_and_convert_llama(convert_args: ConvertArguments, quantization_args
 
     # Call the function with parsed arguments
     download_and_convert_llama(convert_args, quantization_args)
+
+
diff --git a/llava/model/coat/activation/models/coat_olmo.py b/llava/model/coat/activation/models/coat_olmo.py
index 532e4853..64c8496e 100644
--- a/llava/model/coat/activation/models/coat_olmo.py
+++ b/llava/model/coat/activation/models/coat_olmo.py
@@ -1934,3 +1934,5 @@ def _make_state_dict_compatible(
             og_keys_to_new[og_key].add(new_key)
 
         return state_dict, og_keys_to_new
+
+
diff --git a/llava/model/coat/activation/real_quantization/__init__.py b/llava/model/coat/activation/real_quantization/__init__.py
index 2da44200..b33ee32a 100644
--- a/llava/model/coat/activation/real_quantization/__init__.py
+++ b/llava/model/coat/activation/real_quantization/__init__.py
@@ -23,3 +23,5 @@
 from .mul_fwd import fp8_mul_forward
 from .silu_bwd import fp8_silu_backward
 from .silu_fwd import fp8_silu_forward
+
+
diff --git a/llava/model/coat/activation/real_quantization/_dequantize.py b/llava/model/coat/activation/real_quantization/_dequantize.py
index 42171249..b94ae4d4 100644
--- a/llava/model/coat/activation/real_quantization/_dequantize.py
+++ b/llava/model/coat/activation/real_quantization/_dequantize.py
@@ -154,3 +154,5 @@ def fp8_dequantize(x, s_x, QB):
         y = y.reshape(BS, -1, y.shape[-1])
 
     return y
+
+
diff --git a/llava/model/coat/activation/real_quantization/_division.py b/llava/model/coat/activation/real_quantization/_division.py
index 5876dc5f..a22b37f7 100644
--- a/llava/model/coat/activation/real_quantization/_division.py
+++ b/llava/model/coat/activation/real_quantization/_division.py
@@ -204,3 +204,5 @@ def fp8_division(x, QB, fp8type, s_y=None, stochastic=False):
         y = y.reshape(BS, -1, y.shape[-1])
 
     return y, s_y  # y_t is expected to be 2D tensor
+
+
diff --git a/llava/model/coat/activation/real_quantization/_division_transpose.py b/llava/model/coat/activation/real_quantization/_division_transpose.py
index 2297b91d..4165397d 100644
--- a/llava/model/coat/activation/real_quantization/_division_transpose.py
+++ b/llava/model/coat/activation/real_quantization/_division_transpose.py
@@ -207,3 +207,5 @@ def fp8_division_transpose(x, QB, fp8type, s_y=None, stochastic=False, only_tran
         return y, s_y, y_t  # y_t is expected to be 2D tensor
     else:
         return y_t, s_y
+
+
diff --git a/llava/model/coat/activation/real_quantization/_memory_io.py b/llava/model/coat/activation/real_quantization/_memory_io.py
index 480da9c7..d12d2bc3 100644
--- a/llava/model/coat/activation/real_quantization/_memory_io.py
+++ b/llava/model/coat/activation/real_quantization/_memory_io.py
@@ -172,3 +172,5 @@ def y_fwd():
     torch.manual_seed(0)
     torch.set_printoptions(precision=8, linewidth=1600, sci_mode=False, edgeitems=3)
     bench_load_store.run(print_data=True)
+
+
diff --git a/llava/model/coat/activation/real_quantization/_quantize.py b/llava/model/coat/activation/real_quantization/_quantize.py
index aa441fbd..f7e3f45a 100644
--- a/llava/model/coat/activation/real_quantization/_quantize.py
+++ b/llava/model/coat/activation/real_quantization/_quantize.py
@@ -168,3 +168,5 @@ def fp8_quantize(x, QB, fp8type):
         s_y = s_y.reshape(BS, -1, s_y.shape[-1])
 
     return y, s_y
+
+
diff --git a/llava/model/coat/activation/real_quantization/_quantize_pertensor.py b/llava/model/coat/activation/real_quantization/_quantize_pertensor.py
index 66595dd1..1a55ed14 100644
--- a/llava/model/coat/activation/real_quantization/_quantize_pertensor.py
+++ b/llava/model/coat/activation/real_quantization/_quantize_pertensor.py
@@ -144,3 +144,5 @@ def fp8_quantize_pertensor(x, QB, fp8type, stochastic=False):
         s_y = s_y.reshape(BS, -1, s_y.shape[-1])
 
     return y, s_y_max, s_y
+
+
diff --git a/llava/model/coat/activation/real_quantization/_quantize_pertensor_transpose.py b/llava/model/coat/activation/real_quantization/_quantize_pertensor_transpose.py
index b410bece..2de67dda 100644
--- a/llava/model/coat/activation/real_quantization/_quantize_pertensor_transpose.py
+++ b/llava/model/coat/activation/real_quantization/_quantize_pertensor_transpose.py
@@ -147,3 +147,5 @@ def fp8_quantize_pertensor_transpose(x, QB, fp8type, transpose_output_2d=False,
             qy_t = qy_t.reshape(BS, -1, qy_t.shape[-1])
 
     return qy, s_y_max, qy_t  # y_t is expected to be 2D tensor
+
+
diff --git a/llava/model/coat/activation/real_quantization/_transpose.py b/llava/model/coat/activation/real_quantization/_transpose.py
index d27777c4..c1fb878e 100644
--- a/llava/model/coat/activation/real_quantization/_transpose.py
+++ b/llava/model/coat/activation/real_quantization/_transpose.py
@@ -113,3 +113,5 @@ def fp8_transpose(x, transpose_output_2d=False):
         y = y.reshape(BS, -1, y.shape[-1])
 
     return y
+
+
diff --git a/llava/model/coat/activation/real_quantization/add_bwd.py b/llava/model/coat/activation/real_quantization/add_bwd.py
index 7201000a..822f4644 100644
--- a/llava/model/coat/activation/real_quantization/add_bwd.py
+++ b/llava/model/coat/activation/real_quantization/add_bwd.py
@@ -197,3 +197,5 @@ def fp8_add_Ifp_Ifp_Ofp_Opt(x1, x2, QB, fp8type, stochastic=False):  # suppose x
         s_y2 = s_y2.reshape(BS, -1, s_y2.shape[-1])
 
     return y1, (qy2, s_y2_max, s_y2)
+
+
diff --git a/llava/model/coat/activation/real_quantization/add_fwd.py b/llava/model/coat/activation/real_quantization/add_fwd.py
index 254cb385..98c6ab1d 100644
--- a/llava/model/coat/activation/real_quantization/add_fwd.py
+++ b/llava/model/coat/activation/real_quantization/add_fwd.py
@@ -211,3 +211,5 @@ def fp8_add_Ifp_Ifp_Ofp_Og16(x1, x2, fp8type, QB):  # suppose x1 is full precisi
         s_y2 = s_y2.reshape(BS, -1, s_y2.shape[-1])
 
     return y1, (y2, s_y2)
+
+
diff --git a/llava/model/coat/activation/real_quantization/common.py b/llava/model/coat/activation/real_quantization/common.py
index 291e12a2..583c2449 100644
--- a/llava/model/coat/activation/real_quantization/common.py
+++ b/llava/model/coat/activation/real_quantization/common.py
@@ -51,3 +51,5 @@ def get_configs_io_block():
 #                     SCALE_MIN_THRES: tl.constexpr,
 #  + SCALE_MIN_THRES
 # SCALE_MIN_THRES=SCALE_MIN_THRES,
+
+
diff --git a/llava/model/coat/activation/real_quantization/fp8linear.py b/llava/model/coat/activation/real_quantization/fp8linear.py
index 60fee9ce..814f512e 100644
--- a/llava/model/coat/activation/real_quantization/fp8linear.py
+++ b/llava/model/coat/activation/real_quantization/fp8linear.py
@@ -242,3 +242,5 @@ def backward(ctx, grad_output):
                 )
 
         return grad_input, grad_weight, grad_bias, None, None
+
+
diff --git a/llava/model/coat/activation/real_quantization/func_layernorm_noparam.py b/llava/model/coat/activation/real_quantization/func_layernorm_noparam.py
index 19eb9f9a..5aaa1894 100644
--- a/llava/model/coat/activation/real_quantization/func_layernorm_noparam.py
+++ b/llava/model/coat/activation/real_quantization/func_layernorm_noparam.py
@@ -295,3 +295,5 @@ def fp8_layernorm_noparam_backward(x, s_x, g, QB, m, v, num_warps, stochastic=Fa
         dx = dx.reshape(BS, -1, dx.shape[-1])
 
     return dx
+
+
diff --git a/llava/model/coat/activation/real_quantization/func_quantize.py b/llava/model/coat/activation/real_quantization/func_quantize.py
index 517d7c3e..820971a4 100644
--- a/llava/model/coat/activation/real_quantization/func_quantize.py
+++ b/llava/model/coat/activation/real_quantization/func_quantize.py
@@ -94,3 +94,5 @@ def backward(ctx, grad_output):
         Qgrad_output = Qgrad_output.view(torch.float8_e4m3fn)
 
         return grad_output, Qgrad_output, Gscale_g16, None, None
+
+
diff --git a/llava/model/coat/activation/real_quantization/func_rmsnorm.py b/llava/model/coat/activation/real_quantization/func_rmsnorm.py
index 7337dfb7..34aac768 100644
--- a/llava/model/coat/activation/real_quantization/func_rmsnorm.py
+++ b/llava/model/coat/activation/real_quantization/func_rmsnorm.py
@@ -337,3 +337,5 @@ def fp8_rmsnorm_backward(x, s_x, g, w, v, QB, num_warps):
     _rms_norm_bwd_dwdb[grid](_dw, dw, min(GROUP_SIZE_M, M), N, BLOCK_SIZE_M=32, BLOCK_SIZE_N=128, num_ctas=1)  #  #
 
     return dx, dw
+
+
diff --git a/llava/model/coat/activation/real_quantization/gelu_bwd.py b/llava/model/coat/activation/real_quantization/gelu_bwd.py
index dacd8e92..3171cf22 100644
--- a/llava/model/coat/activation/real_quantization/gelu_bwd.py
+++ b/llava/model/coat/activation/real_quantization/gelu_bwd.py
@@ -227,3 +227,5 @@ def fp8_gelu_backward(x, s_x, g, QB, fp8type):
         s_y = s_y.reshape(BS, -1, s_y.shape[-1])
 
     return qy, s_y_max
+
+
diff --git a/llava/model/coat/activation/real_quantization/gelu_bwd_legacy.py b/llava/model/coat/activation/real_quantization/gelu_bwd_legacy.py
index e301385f..9d1cb114 100644
--- a/llava/model/coat/activation/real_quantization/gelu_bwd_legacy.py
+++ b/llava/model/coat/activation/real_quantization/gelu_bwd_legacy.py
@@ -249,3 +249,5 @@ def fp8_gelu_backward_legacy(x, s_x, g, s_g, QB):
         s_y = s_y.reshape(BS, -1, s_y.shape[-1])
 
     return qy, s_y_max
+
+
diff --git a/llava/model/coat/activation/real_quantization/gelu_fwd.py b/llava/model/coat/activation/real_quantization/gelu_fwd.py
index 48ab20d0..0aa7b524 100644
--- a/llava/model/coat/activation/real_quantization/gelu_fwd.py
+++ b/llava/model/coat/activation/real_quantization/gelu_fwd.py
@@ -201,3 +201,5 @@ def fp8_gelu_forward(x, s_x, QB, transpose_output_2d=False):
             qy_t = qy_t.reshape(BS, -1, qy.shape[-1])
 
     return qy, s_y_max, qy_t
+
+
diff --git a/llava/model/coat/activation/real_quantization/linear.py b/llava/model/coat/activation/real_quantization/linear.py
index 756e1827..c2239d88 100644
--- a/llava/model/coat/activation/real_quantization/linear.py
+++ b/llava/model/coat/activation/real_quantization/linear.py
@@ -299,3 +299,5 @@ def fp8_linear_backward(
         return y, s_y, w_g
     else:
         return y, w_g
+
+
diff --git a/llava/model/coat/activation/real_quantization/mul_bwd.py b/llava/model/coat/activation/real_quantization/mul_bwd.py
index 63e6201d..247bb1d2 100644
--- a/llava/model/coat/activation/real_quantization/mul_bwd.py
+++ b/llava/model/coat/activation/real_quantization/mul_bwd.py
@@ -312,3 +312,5 @@ def fp8_mul_backward(
             s_y2 = s_y2.reshape(BS, -1, s_y2.shape[-1])
 
         return y1, (qy2, s_y2_max, qy2_t)
+
+
diff --git a/llava/model/coat/activation/real_quantization/mul_bwd_legacy.py b/llava/model/coat/activation/real_quantization/mul_bwd_legacy.py
index 9e4f0646..4039e8e4 100644
--- a/llava/model/coat/activation/real_quantization/mul_bwd_legacy.py
+++ b/llava/model/coat/activation/real_quantization/mul_bwd_legacy.py
@@ -366,3 +366,5 @@ def fp8_mul_backward_legacy(
         s_y2 = s_y2.reshape(BS, -1, s_y2.shape[-1])
 
     return y1, s_y1, y2, s_y2
+
+
diff --git a/llava/model/coat/activation/real_quantization/mul_bwd_silu_fwd.py b/llava/model/coat/activation/real_quantization/mul_bwd_silu_fwd.py
index c469a872..b8be7370 100644
--- a/llava/model/coat/activation/real_quantization/mul_bwd_silu_fwd.py
+++ b/llava/model/coat/activation/real_quantization/mul_bwd_silu_fwd.py
@@ -329,3 +329,5 @@ def fp8_mul_backward_silu_forward(
             s_y2 = s_y2.reshape(BS, -1, s_y2.shape[-1])
 
         return y1, (qy2, s_y2_max, qy2_t)
+
+
diff --git a/llava/model/coat/activation/real_quantization/mul_fwd.py b/llava/model/coat/activation/real_quantization/mul_fwd.py
index a0ddd21c..32c0e7b6 100644
--- a/llava/model/coat/activation/real_quantization/mul_fwd.py
+++ b/llava/model/coat/activation/real_quantization/mul_fwd.py
@@ -252,3 +252,5 @@ def fp8_mul_forward(x1, s_x1, x2, s_x2, QB, transpose_output_2d=False):
             qy_t = qy_t.reshape(BS, -1, qy.shape[-1])
 
     return qy, s_y_max, qy_t
+
+
diff --git a/llava/model/coat/activation/real_quantization/silu_bwd.py b/llava/model/coat/activation/real_quantization/silu_bwd.py
index a8ed684f..11189ed9 100644
--- a/llava/model/coat/activation/real_quantization/silu_bwd.py
+++ b/llava/model/coat/activation/real_quantization/silu_bwd.py
@@ -247,3 +247,5 @@ def fp8_silu_backward(
     #     s_y = s_y.reshape(BS, -1, s_y.shape[-1])
 
     # return qy, s_y_max
+
+
diff --git a/llava/model/coat/activation/real_quantization/silu_bwd_legacy.py b/llava/model/coat/activation/real_quantization/silu_bwd_legacy.py
index 009d9910..a90ad412 100644
--- a/llava/model/coat/activation/real_quantization/silu_bwd_legacy.py
+++ b/llava/model/coat/activation/real_quantization/silu_bwd_legacy.py
@@ -240,3 +240,5 @@ def fp8_silu_backward_legacy(x, s_x, g, s_g, QB, stochastic=False):  # Stochasti
         s_y = s_y.reshape(BS, -1, s_y.shape[-1])
 
     return y, s_y
+
+
diff --git a/llava/model/coat/activation/real_quantization/silu_fwd.py b/llava/model/coat/activation/real_quantization/silu_fwd.py
index 2168aa36..5738cb62 100644
--- a/llava/model/coat/activation/real_quantization/silu_fwd.py
+++ b/llava/model/coat/activation/real_quantization/silu_fwd.py
@@ -196,3 +196,5 @@ def fp8_silu_forward(x, s_x, QB):
         s_y = s_y.reshape(BS, -1, s_y.shape[-1])
 
     return y, s_y
+
+
diff --git a/llava/model/coat/activation/utils.py b/llava/model/coat/activation/utils.py
index 9dcd3c16..66df713a 100644
--- a/llava/model/coat/activation/utils.py
+++ b/llava/model/coat/activation/utils.py
@@ -25,3 +25,5 @@ def quant_get_local_rank() -> int:
 
 
 record_memory_allocated = defaultdict(list)
+
+
diff --git a/llava/model/coat/fp8_trainer.py b/llava/model/coat/fp8_trainer.py
index 9edb8b38..f632b228 100644
--- a/llava/model/coat/fp8_trainer.py
+++ b/llava/model/coat/fp8_trainer.py
@@ -618,3 +618,5 @@ def _inner_training_loop(
             self._deactivate_neftune(self.model)
 
         return TrainOutput(self.state.global_step, train_loss, metrics)
+
+
diff --git a/llava/model/coat/optimizer/fp8_adamw.py b/llava/model/coat/optimizer/fp8_adamw.py
index 50b0423e..5030cc9b 100644
--- a/llava/model/coat/optimizer/fp8_adamw.py
+++ b/llava/model/coat/optimizer/fp8_adamw.py
@@ -507,3 +507,5 @@ def _single_tensor_Coatadamw(
                 step,
                 qgroup_size,
             )
+
+
diff --git a/llava/model/coat/optimizer/kernels/setup.py b/llava/model/coat/optimizer/kernels/setup.py
index a6c016fa..a55fdb8a 100644
--- a/llava/model/coat/optimizer/kernels/setup.py
+++ b/llava/model/coat/optimizer/kernels/setup.py
@@ -48,3 +48,5 @@
     ],
     cmdclass={"build_ext": BuildExtension},
 )
+
+
diff --git a/llava/model/configuration_llava.py b/llava/model/configuration_llava.py
index d83008b3..6d43fc7c 100755
--- a/llava/model/configuration_llava.py
+++ b/llava/model/configuration_llava.py
@@ -87,3 +87,5 @@ def __init__(
 
         self.image_encoder = image_encoder
         self.video_encoder = video_encoder
+
+
diff --git a/llava/model/consolidate.py b/llava/model/consolidate.py
index 2e86ed17..e9014bc0 100755
--- a/llava/model/consolidate.py
+++ b/llava/model/consolidate.py
@@ -44,3 +44,5 @@ def consolidate_ckpt(src_path, dst_path):
     args = parser.parse_args()
 
     consolidate_ckpt(args.src, args.dst)
+
+
diff --git a/llava/model/encoders/__init__.py b/llava/model/encoders/__init__.py
index 51656ca5..df727be2 100644
--- a/llava/model/encoders/__init__.py
+++ b/llava/model/encoders/__init__.py
@@ -1,2 +1,4 @@
 from .image import *
 from .video import *
+
+
diff --git a/llava/model/encoders/base.py b/llava/model/encoders/base.py
index 00a62e76..3ce82f0a 100644
--- a/llava/model/encoders/base.py
+++ b/llava/model/encoders/base.py
@@ -11,3 +11,5 @@ def __init__(self, parent: nn.Module) -> None:
     @property
     def parent(self) -> nn.Module:
         return self._parent[0]
+
+
diff --git a/llava/model/encoders/image/__init__.py b/llava/model/encoders/image/__init__.py
index da0f2b60..a480aa21 100644
--- a/llava/model/encoders/image/__init__.py
+++ b/llava/model/encoders/image/__init__.py
@@ -1 +1,3 @@
 from .basic import *
+
+
diff --git a/llava/model/encoders/image/basic.py b/llava/model/encoders/image/basic.py
index fa25b640..00b981d5 100644
--- a/llava/model/encoders/image/basic.py
+++ b/llava/model/encoders/image/basic.py
@@ -47,3 +47,5 @@ def forward(self, images: List[torch.Tensor], config: Dict[str, Any]) -> List[to
             end_token_embeds=self.embed_tokens(self.end_tokens),
         )
         return [process_features(f) for f in features]
+
+
diff --git a/llava/model/encoders/video/__init__.py b/llava/model/encoders/video/__init__.py
index 7599d595..1a375f29 100644
--- a/llava/model/encoders/video/__init__.py
+++ b/llava/model/encoders/video/__init__.py
@@ -1,2 +1,4 @@
 from .basic import *
 from .tsp import *
+
+
diff --git a/llava/model/encoders/video/basic.py b/llava/model/encoders/video/basic.py
index 6ef5f9f6..8fa01180 100644
--- a/llava/model/encoders/video/basic.py
+++ b/llava/model/encoders/video/basic.py
@@ -51,3 +51,5 @@ def forward(self, videos: List[torch.Tensor], config: Dict[str, Any]) -> List[to
             end_token_embeds=self.embed_tokens(self.end_tokens),
         )
         return [process_features(f) for f in features]
+
+
diff --git a/llava/model/encoders/video/tsp.py b/llava/model/encoders/video/tsp.py
index db4f3863..68c39232 100644
--- a/llava/model/encoders/video/tsp.py
+++ b/llava/model/encoders/video/tsp.py
@@ -62,3 +62,5 @@ def forward(self, videos: List[torch.Tensor], config: Dict[str, Any]) -> List[to
             sep_token_embeds=self.embed_tokens(self.sep_tokens),
         )
         return [process_features(f) for f in features]
+
+
diff --git a/llava/model/language_model/builder.py b/llava/model/language_model/builder.py
index 6d675f44..e20f4f17 100755
--- a/llava/model/language_model/builder.py
+++ b/llava/model/language_model/builder.py
@@ -213,3 +213,5 @@ def build_llm_and_tokenizer(
     # TODO(ligeng): is this necessary for llava?
     config.hidden_size = llm.config.hidden_size
     return llm, tokenizer
+
+
diff --git a/llava/model/language_model/configuration_quantize.py b/llava/model/language_model/configuration_quantize.py
index ca997c76..08de2a6c 100644
--- a/llava/model/language_model/configuration_quantize.py
+++ b/llava/model/language_model/configuration_quantize.py
@@ -69,3 +69,5 @@ def __init__(
 #         self.col_blocksize = col_blocksize
 #         self.qchoice = qchoice
 #         self.pad_to_multiple_of = pad_to_multiple_of
+
+
diff --git a/llava/model/language_model/fp8_qwen2_convert_from_hf.py b/llava/model/language_model/fp8_qwen2_convert_from_hf.py
index f04563b3..77915f82 100644
--- a/llava/model/language_model/fp8_qwen2_convert_from_hf.py
+++ b/llava/model/language_model/fp8_qwen2_convert_from_hf.py
@@ -65,3 +65,5 @@ def download_and_convert_qwen2(convert_args: ConvertArguments, quantization_args
 
     # Call the function with parsed arguments
     download_and_convert_qwen2(convert_args, quantization_args)
+
+
diff --git a/llava/model/language_model/fp8activationqwen2.py b/llava/model/language_model/fp8activationqwen2.py
index d17e0a56..84ea813e 100644
--- a/llava/model/language_model/fp8activationqwen2.py
+++ b/llava/model/language_model/fp8activationqwen2.py
@@ -1714,3 +1714,5 @@ def make_state_dict_compatible(state_dict: dict[str, torch.Tensor]):
         compatible_state_dict[new_key] = value
 
     return compatible_state_dict
+
+
diff --git a/llava/model/language_model/fp8activationresidualqwen2.py b/llava/model/language_model/fp8activationresidualqwen2.py
index e3cd3a63..c84cac7c 100644
--- a/llava/model/language_model/fp8activationresidualqwen2.py
+++ b/llava/model/language_model/fp8activationresidualqwen2.py
@@ -1578,3 +1578,5 @@ def make_state_dict_compatible(state_dict: dict[str, torch.Tensor]):
         compatible_state_dict[new_key] = value
 
     return compatible_state_dict
+
+
diff --git a/llava/model/language_model/fp8linearqwen2.py b/llava/model/language_model/fp8linearqwen2.py
index a302fa5d..312b095a 100644
--- a/llava/model/language_model/fp8linearqwen2.py
+++ b/llava/model/language_model/fp8linearqwen2.py
@@ -348,3 +348,5 @@ def loss_function(self):
 AutoConfig.register("fp8linear_qwen2", FP8LinearQwen2Config)
 AutoModel.register(FP8LinearQwen2Config, FP8LinearQwen2Model)
 AutoModelForCausalLM.register(FP8LinearQwen2Config, FP8LinearQwen2ForCausalLM)
+
+
diff --git a/llava/model/language_model/llava_llama.py b/llava/model/language_model/llava_llama.py
index afd06fdd..ac4a0132 100755
--- a/llava/model/language_model/llava_llama.py
+++ b/llava/model/language_model/llava_llama.py
@@ -160,3 +160,5 @@ def forward(
 
 AutoConfig.register("llava_llama", LlavaLlamaConfig)
 AutoModel.register(LlavaLlamaConfig, LlavaLlamaModel)
+
+
diff --git a/llava/model/language_model/qllama.py b/llava/model/language_model/qllama.py
index 11162566..112c70ad 100644
--- a/llava/model/language_model/qllama.py
+++ b/llava/model/language_model/qllama.py
@@ -304,3 +304,5 @@ def set_input_embeddings(self, value):
 AutoConfig.register("qllama", QLlamaConfig)
 AutoModel.register(QLlamaConfig, QLlamaModel)
 AutoModelForCausalLM.register(QLlamaConfig, QLlamaForCausalLM)
+
+
diff --git a/llava/model/language_model/qllava_qllama.py b/llava/model/language_model/qllava_qllama.py
index 4aa42e3b..4c717cd7 100755
--- a/llava/model/language_model/qllava_qllama.py
+++ b/llava/model/language_model/qllava_qllama.py
@@ -114,3 +114,5 @@ def init_vlm(self, config: PreTrainedModel = None, model_args=None, *args, **kwa
 
 AutoConfig.register("qllava_qllama", QLlavaLlamaConfig)
 AutoModel.register(QLlavaLlamaConfig, QLlavaLlamaModel)
+
+
diff --git a/llava/model/language_model/qmemllama.py b/llava/model/language_model/qmemllama.py
index 6f221b77..8829538c 100644
--- a/llava/model/language_model/qmemllama.py
+++ b/llava/model/language_model/qmemllama.py
@@ -671,3 +671,5 @@ def set_input_embeddings(self, value):
 AutoConfig.register("qmemllama", QMemLlamaConfig)
 AutoModel.register(QMemLlamaConfig, QMemLlamaModel)
 AutoModelForCausalLM.register(QMemLlamaConfig, QMemLlamaForCausalLM)
+
+
diff --git a/llava/model/language_model/realqmemllama.py b/llava/model/language_model/realqmemllama.py
index a90664b8..ec66a842 100644
--- a/llava/model/language_model/realqmemllama.py
+++ b/llava/model/language_model/realqmemllama.py
@@ -666,3 +666,5 @@ def set_input_embeddings(self, value):
 AutoConfig.register("qmemllama", QMemLlamaConfig)
 AutoModel.register(QMemLlamaConfig, QMemLlamaModel)
 AutoModelForCausalLM.register(QMemLlamaConfig, QMemLlamaForCausalLM)
+
+
diff --git a/llava/model/liger/cross_entropy.py b/llava/model/liger/cross_entropy.py
index a95c0c11..cbb67b73 100644
--- a/llava/model/liger/cross_entropy.py
+++ b/llava/model/liger/cross_entropy.py
@@ -409,3 +409,5 @@ def LigerForCausalLMLoss(
     shift_labels = shift_labels.to(shift_logits.device)
     loss = liger_fixed_cross_entropy(shift_logits, shift_labels, num_items_in_batch, ignore_index, **kwargs)
     return loss
+
+
diff --git a/llava/model/liger/utils.py b/llava/model/liger/utils.py
index 4a24223d..89202d48 100644
--- a/llava/model/liger/utils.py
+++ b/llava/model/liger/utils.py
@@ -121,3 +121,5 @@ def element_mul_kernel(
         X_offsets = i + tl.arange(0, BLOCK_SIZE)
         X_block = tl.load(X_ptr + X_offsets, mask=X_offsets < n_cols)
         tl.store(X_ptr + X_offsets, X_block * grad_output, mask=X_offsets < n_cols)
+
+
diff --git a/llava/model/llava_arch.py b/llava/model/llava_arch.py
index 60115d53..b682aeef 100755
--- a/llava/model/llava_arch.py
+++ b/llava/model/llava_arch.py
@@ -866,3 +866,5 @@ def default_generation_config(self) -> GenerationConfig:
         if generation_config.eos_token_id is None:
             generation_config.eos_token_id = self.tokenizer.stop_token_ids
         return generation_config
+
+
diff --git a/llava/model/loss.py b/llava/model/loss.py
index 47d683cd..4ba74c41 100644
--- a/llava/model/loss.py
+++ b/llava/model/loss.py
@@ -46,3 +46,5 @@ def soft_cross_entropy(
 
     # Return average loss
     return loss / targets.size(0)
+
+
diff --git a/llava/model/make_delta.py b/llava/model/make_delta.py
index 5a18327b..90b948d8 100755
--- a/llava/model/make_delta.py
+++ b/llava/model/make_delta.py
@@ -74,3 +74,5 @@ def make_delta(base_model_path, target_model_path, delta_path, hub_repo_id):
     args = parser.parse_args()
 
     make_delta(args.base_model_path, args.target_model_path, args.delta_path, args.hub_repo_id)
+
+
diff --git a/llava/model/multimodal_encoder/builder.py b/llava/model/multimodal_encoder/builder.py
index e784935a..1c7acfc7 100755
--- a/llava/model/multimodal_encoder/builder.py
+++ b/llava/model/multimodal_encoder/builder.py
@@ -68,3 +68,5 @@ def build_vision_tower(model_name_or_path: str, config: PretrainedConfig) -> Pre
         vision_tower.config.hidden_size if not (use_s2 or use_dynamic_s2) else vision_tower.hidden_size
     )
     return vision_tower
+
+
diff --git a/llava/model/multimodal_encoder/clip_encoder.py b/llava/model/multimodal_encoder/clip_encoder.py
index a6c4c45b..189db8c4 100755
--- a/llava/model/multimodal_encoder/clip_encoder.py
+++ b/llava/model/multimodal_encoder/clip_encoder.py
@@ -40,3 +40,5 @@ def __init__(self, model_name_or_path: str, config: PretrainedConfig):
         self.image_processor.crop_size["height"] = self.image_processor.crop_size["width"] = self.scales[-1]
 
         self.is_loaded = True
+
+
diff --git a/llava/model/multimodal_encoder/image_processor.py b/llava/model/multimodal_encoder/image_processor.py
index 393d29ec..c21381dd 100755
--- a/llava/model/multimodal_encoder/image_processor.py
+++ b/llava/model/multimodal_encoder/image_processor.py
@@ -544,3 +544,5 @@ def preprocess(
         }
 
         return BatchFeature(data=data, tensor_type=return_tensors)
+
+
diff --git a/llava/model/multimodal_encoder/intern/configuration_intern_vit.py b/llava/model/multimodal_encoder/intern/configuration_intern_vit.py
index c8b49d55..4b15c4fe 100755
--- a/llava/model/multimodal_encoder/intern/configuration_intern_vit.py
+++ b/llava/model/multimodal_encoder/intern/configuration_intern_vit.py
@@ -115,3 +115,5 @@ def from_pretrained(cls, pretrained_model_name_or_path: Union[str, os.PathLike],
             )
 
         return cls.from_dict(config_dict, **kwargs)
+
+
diff --git a/llava/model/multimodal_encoder/intern/flash_attention.py b/llava/model/multimodal_encoder/intern/flash_attention.py
index cf5b162f..05eaaa97 100755
--- a/llava/model/multimodal_encoder/intern/flash_attention.py
+++ b/llava/model/multimodal_encoder/intern/flash_attention.py
@@ -103,3 +103,5 @@ def forward(self, qkv, key_padding_mask=None, causal=False, cu_seqlens=None, max
             )
 
         return output, None
+
+
diff --git a/llava/model/multimodal_encoder/intern/modeling_intern_vit.py b/llava/model/multimodal_encoder/intern/modeling_intern_vit.py
index 40cc3e82..b38f7c6e 100755
--- a/llava/model/multimodal_encoder/intern/modeling_intern_vit.py
+++ b/llava/model/multimodal_encoder/intern/modeling_intern_vit.py
@@ -541,3 +541,5 @@ def forward(
             hidden_states=encoder_outputs.hidden_states,
             attentions=encoder_outputs.attentions,
         )
+
+
diff --git a/llava/model/multimodal_encoder/intern_encoder.py b/llava/model/multimodal_encoder/intern_encoder.py
index 3645a168..118924f2 100755
--- a/llava/model/multimodal_encoder/intern_encoder.py
+++ b/llava/model/multimodal_encoder/intern_encoder.py
@@ -88,3 +88,5 @@ def __init__(self, vision_tower, config, drop_path_rate=0.0):
 
 AutoConfig.register("intern_vit_6b", InternVisionConfig)
 AutoModel.register(InternVisionConfig, InternVisionModel)
+
+
diff --git a/llava/model/multimodal_encoder/radio_encoder.py b/llava/model/multimodal_encoder/radio_encoder.py
index 838eb392..089d9887 100755
--- a/llava/model/multimodal_encoder/radio_encoder.py
+++ b/llava/model/multimodal_encoder/radio_encoder.py
@@ -332,3 +332,5 @@ def forward(self, images: torch.Tensor):
         self.sample_count += 1
 
         return features
+
+
diff --git a/llava/model/multimodal_encoder/radio_torchhub_encoder.py b/llava/model/multimodal_encoder/radio_torchhub_encoder.py
index 05c07e7f..c97f01bc 100644
--- a/llava/model/multimodal_encoder/radio_torchhub_encoder.py
+++ b/llava/model/multimodal_encoder/radio_torchhub_encoder.py
@@ -373,3 +373,5 @@ def forward(self, images: torch.Tensor):
         self.sample_count += 1
 
         return features
+
+
diff --git a/llava/model/multimodal_encoder/siglip/__init__.py b/llava/model/multimodal_encoder/siglip/__init__.py
index 0e44734b..f0f851ad 100644
--- a/llava/model/multimodal_encoder/siglip/__init__.py
+++ b/llava/model/multimodal_encoder/siglip/__init__.py
@@ -1 +1,3 @@
 from .modeling_siglip import SiglipVisionModel
+
+
diff --git a/llava/model/multimodal_encoder/siglip/modeling_siglip.py b/llava/model/multimodal_encoder/siglip/modeling_siglip.py
index a4339645..231af138 100644
--- a/llava/model/multimodal_encoder/siglip/modeling_siglip.py
+++ b/llava/model/multimodal_encoder/siglip/modeling_siglip.py
@@ -1669,3 +1669,5 @@ def forward(
             hidden_states=outputs.hidden_states,
             attentions=outputs.attentions,
         )
+
+
diff --git a/llava/model/multimodal_encoder/siglip_encoder.py b/llava/model/multimodal_encoder/siglip_encoder.py
index da11ae32..ddb73833 100755
--- a/llava/model/multimodal_encoder/siglip_encoder.py
+++ b/llava/model/multimodal_encoder/siglip_encoder.py
@@ -61,3 +61,5 @@ def __init__(self, model_name_or_path: str, config: PretrainedConfig) -> None:
         # Make sure it crops/resizes the image to the largest scale in self.scales to maintain high-res information
         self.image_processor.size["height"] = self.image_processor.size["width"] = self.scales[0]
         self.is_loaded = True
+
+
diff --git a/llava/model/multimodal_encoder/vision_encoder.py b/llava/model/multimodal_encoder/vision_encoder.py
index 80a48072..fd04df53 100755
--- a/llava/model/multimodal_encoder/vision_encoder.py
+++ b/llava/model/multimodal_encoder/vision_encoder.py
@@ -247,3 +247,5 @@ def forward(self, images):
     @property
     def hidden_size(self):
         return self.config.hidden_size * len(self.scales)
+
+
diff --git a/llava/model/multimodal_encoder/visualize_features.py b/llava/model/multimodal_encoder/visualize_features.py
index e9b87f24..6e09d204 100755
--- a/llava/model/multimodal_encoder/visualize_features.py
+++ b/llava/model/multimodal_encoder/visualize_features.py
@@ -349,3 +349,5 @@ def get_cluster_map(
     #     world_size = dist.get_world_size()
 
     main(rank, world_size)
+
+
diff --git a/llava/model/multimodal_projector/base_projector.py b/llava/model/multimodal_projector/base_projector.py
index f94813d6..5cbb0d88 100755
--- a/llava/model/multimodal_projector/base_projector.py
+++ b/llava/model/multimodal_projector/base_projector.py
@@ -226,3 +226,5 @@ def forward(self, x, *args, **kwargs):
 
 AutoConfig.register("v2l_projector", MultimodalProjectorConfig)
 AutoModel.register(MultimodalProjectorConfig, MultimodalProjector)
+
+
diff --git a/llava/model/multimodal_projector/builder.py b/llava/model/multimodal_projector/builder.py
index 4fb0c832..615906e7 100755
--- a/llava/model/multimodal_projector/builder.py
+++ b/llava/model/multimodal_projector/builder.py
@@ -37,3 +37,5 @@ def build_mm_projector(model_type_or_path: str, config: PretrainedConfig) -> Pre
         mm_projector_cfg = MultimodalProjectorConfig(model_type_or_path)
         mm_projector = MultimodalProjector(mm_projector_cfg, config).to(eval(config.model_dtype))
         return mm_projector
+
+
diff --git a/llava/model/qfunction.py b/llava/model/qfunction.py
index 816afdf9..b755b783 100644
--- a/llava/model/qfunction.py
+++ b/llava/model/qfunction.py
@@ -272,3 +272,5 @@ def backward(ctx, grad_output):
     # Q_X = block_quant(B_X, symm=True, bits=2, stochastic=False)
     # print(Q_X)
     # print(Q_X.shape)
+
+
diff --git a/llava/model/qlinear_te.py b/llava/model/qlinear_te.py
index 36fd09cb..79704dea 100644
--- a/llava/model/qlinear_te.py
+++ b/llava/model/qlinear_te.py
@@ -223,3 +223,5 @@ def backward(ctx, grad_output):
                 )
 
         return grad_input, grad_weight, grad_bias, None, None
+
+
diff --git a/llava/model/quantization/FloatPointQuantizeTorch.py b/llava/model/quantization/FloatPointQuantizeTorch.py
index 05573449..174792a8 100644
--- a/llava/model/quantization/FloatPointQuantizeTorch.py
+++ b/llava/model/quantization/FloatPointQuantizeTorch.py
@@ -76,3 +76,5 @@ def ZeroDynamic_quantize_torch(x, bit, stochastic):
 
     y = y.to(x)
     return y
+
+
diff --git a/llava/model/quantization/FloatPointQuantizeTriton.py b/llava/model/quantization/FloatPointQuantizeTriton.py
index f6767b8a..145bc613 100644
--- a/llava/model/quantization/FloatPointQuantizeTriton.py
+++ b/llava/model/quantization/FloatPointQuantizeTriton.py
@@ -157,3 +157,5 @@ def _floatExMy_stochastic_quantize_kernel(
     y = y.to(x_ptr.dtype.element_ty)
 
     tl.store(output_ptr + offsets, y, mask=mask)
+
+
diff --git a/llava/model/quantization/QAct.py b/llava/model/quantization/QAct.py
index b1934087..2ddb8ae6 100644
--- a/llava/model/quantization/QAct.py
+++ b/llava/model/quantization/QAct.py
@@ -509,3 +509,5 @@ def backward(ctx, Qgrad_output, Gscale):
     print(if_nan)
 
     Q = block_quant(Binput, True, 8, stochastic=False, epsilon=1e-8)
+
+
diff --git a/llava/model/quantization/QAdd.py b/llava/model/quantization/QAdd.py
index 031f08d2..74684082 100644
--- a/llava/model/quantization/QAdd.py
+++ b/llava/model/quantization/QAdd.py
@@ -56,3 +56,5 @@ def forward(self, Qinput_re, Qinput_fx, Iscale_re, Iscale_fx):
 
 if __name__ == "__main__":
     Sum = torch.load("tensor/QAct_nan_epoch16.pt")
+
+
diff --git a/llava/model/quantization/QFunction.py b/llava/model/quantization/QFunction.py
index 816afdf9..b755b783 100644
--- a/llava/model/quantization/QFunction.py
+++ b/llava/model/quantization/QFunction.py
@@ -272,3 +272,5 @@ def backward(ctx, grad_output):
     # Q_X = block_quant(B_X, symm=True, bits=2, stochastic=False)
     # print(Q_X)
     # print(Q_X.shape)
+
+
diff --git a/llava/model/quantization/QGELU.py b/llava/model/quantization/QGELU.py
index 99fb55d2..2794e2ec 100644
--- a/llava/model/quantization/QGELU.py
+++ b/llava/model/quantization/QGELU.py
@@ -57,3 +57,5 @@ def forward(self, Qinput, Iscale):
 
 if __name__ == "__main__":
     Sum = torch.load("tensor/QAct_nan_epoch16.pt")
+
+
diff --git a/llava/model/quantization/QIdentity.py b/llava/model/quantization/QIdentity.py
index a739cb22..b01845e5 100644
--- a/llava/model/quantization/QIdentity.py
+++ b/llava/model/quantization/QIdentity.py
@@ -54,3 +54,5 @@ def backward(ctx, grad_output, Gscale):
     print(if_nan)
 
     Q = block_quant(Binput, True, 8, stochastic=False, epsilon=1e-8)
+
+
diff --git a/llava/model/quantization/QLayerNorm.py b/llava/model/quantization/QLayerNorm.py
index f84794e8..77e1b080 100644
--- a/llava/model/quantization/QLayerNorm.py
+++ b/llava/model/quantization/QLayerNorm.py
@@ -59,3 +59,5 @@ def forward(self, Qinput, Iscale):
 
 if __name__ == "__main__":
     Sum = torch.load("tensor/QAct_nan_epoch16.pt")
+
+
diff --git a/llava/model/quantization/QLinear.py b/llava/model/quantization/QLinear.py
index 8cf9610c..b51c394c 100644
--- a/llava/model/quantization/QLinear.py
+++ b/llava/model/quantization/QLinear.py
@@ -325,3 +325,5 @@ def backward(ctx, Qgrad_output, Gscale):
             grad_bias = None
 
         return Qgrad_input_transform, GIscale, RQgrad_weight, grad_bias, None, None, None, None, None, None
+
+
diff --git a/llava/model/quantization/QMul.py b/llava/model/quantization/QMul.py
index 8e7e7b6b..0191c880 100644
--- a/llava/model/quantization/QMul.py
+++ b/llava/model/quantization/QMul.py
@@ -59,3 +59,5 @@ def forward(self, Qinput1, Qinput2, Iscale1, Iscale2):
 
 if __name__ == "__main__":
     Sum = torch.load("tensor/QAct_nan_epoch16.pt")
+
+
diff --git a/llava/model/quantization/Qconfig.py b/llava/model/quantization/Qconfig.py
index 8c9db29e..59421365 100644
--- a/llava/model/quantization/Qconfig.py
+++ b/llava/model/quantization/Qconfig.py
@@ -52,3 +52,5 @@ def __init__(self):
 
 
 qconfig = QuantizationConfig()
+
+
diff --git a/llava/model/quantization/__init__.py b/llava/model/quantization/__init__.py
index 028bd28f..20ccbea1 100644
--- a/llava/model/quantization/__init__.py
+++ b/llava/model/quantization/__init__.py
@@ -7,3 +7,5 @@
 from .QLinear import *
 from .QMul import *
 from .utils import *
+
+
diff --git a/llava/model/quantization/utils.py b/llava/model/quantization/utils.py
index 5d348f66..5a62b3cb 100644
--- a/llava/model/quantization/utils.py
+++ b/llava/model/quantization/utils.py
@@ -205,3 +205,5 @@ def visualize_distribution(pt_path):
     print(f"{aw}   {fb}   {layer_name} distribution finish!")
 
     exit(0)
+
+
diff --git a/llava/model/qutils.py b/llava/model/qutils.py
index af7c84e7..d1d201b8 100644
--- a/llava/model/qutils.py
+++ b/llava/model/qutils.py
@@ -203,3 +203,5 @@ def visualize_distribution(pt_path):
     print(f"{aw}   {fb}   {layer_name} distribution finish!")
 
     exit(0)
+
+
diff --git a/llava/model/realquantize/common.py b/llava/model/realquantize/common.py
index 9f818e55..374a68f4 100644
--- a/llava/model/realquantize/common.py
+++ b/llava/model/realquantize/common.py
@@ -17,3 +17,5 @@
 #                     SCALE_MIN_THRES: tl.constexpr,
 #  + SCALE_MIN_THRES
 # SCALE_MIN_THRES=SCALE_MIN_THRES,
+
+
diff --git a/llava/model/realquantize/division.py b/llava/model/realquantize/division.py
index 4af05648..a8d9c5a1 100644
--- a/llava/model/realquantize/division.py
+++ b/llava/model/realquantize/division.py
@@ -280,3 +280,5 @@ def validity_check(BS, SL, CDIM, QB, fp8type=torch.float8_e4m3fn):
     import IPython
 
     IPython.embed()
+
+
diff --git a/llava/model/realquantize/division_transpose.py b/llava/model/realquantize/division_transpose.py
index 4167572b..2c106548 100644
--- a/llava/model/realquantize/division_transpose.py
+++ b/llava/model/realquantize/division_transpose.py
@@ -303,3 +303,5 @@ def validity_check(BS, SL, CDIM, QB, fp8type=torch.float8_e4m3fn):
     import IPython
 
     IPython.embed()
+
+
diff --git a/llava/model/realquantize/linear.py b/llava/model/realquantize/linear.py
index 25351043..0dde2deb 100644
--- a/llava/model/realquantize/linear.py
+++ b/llava/model/realquantize/linear.py
@@ -372,3 +372,5 @@ def benchmark(M, N, K, provider, fp8_inputs):
     # time_check(4096, 11008, 5380)
     # validity_check(2048, 1024, 4096)
     benchmark.run(show_plots=True, print_data=True)
+
+
diff --git a/llava/model/realquantize/quantize_and_transpose.py b/llava/model/realquantize/quantize_and_transpose.py
index 5776331d..66abb714 100644
--- a/llava/model/realquantize/quantize_and_transpose.py
+++ b/llava/model/realquantize/quantize_and_transpose.py
@@ -231,3 +231,5 @@ def validity_check(BS, SL, CDIM, QB, fp8type=torch.float8_e4m3fn):
     torch.set_printoptions(precision=8, linewidth=1600, sci_mode=False, edgeitems=3)
     validity_check(BS=4, SL=256, CDIM=512, QB=16, fp8type=torch.float8_e4m3fn)
     bench_load_store.run(save_path=f"result/time/multi_quantize_block_quantize/BLSZ=64", print_data=True)
+
+
diff --git a/llava/model/realquantize/trans_grad_bias.py b/llava/model/realquantize/trans_grad_bias.py
index a27fe7c8..d1db83ce 100644
--- a/llava/model/realquantize/trans_grad_bias.py
+++ b/llava/model/realquantize/trans_grad_bias.py
@@ -230,3 +230,5 @@ def validity_check(BS, SL, CDIM, QB, fp8type=torch.float8_e4m3fn):
     torch.set_printoptions(precision=8, linewidth=1600, sci_mode=False, edgeitems=3)
     validity_check(BS=4, SL=256, CDIM=512, QB=16, fp8type=torch.float8_e4m3fn)
     bench_load_store.run(save_path=f"result/time/multi_quantize_block_quantize/BLSZ=64", print_data=True)
+
+
diff --git a/llava/model/utils/__init__.py b/llava/model/utils/__init__.py
index 16281fe0..d81436df 100644
--- a/llava/model/utils/__init__.py
+++ b/llava/model/utils/__init__.py
@@ -1 +1,3 @@
 from .utils import *
+
+
diff --git a/llava/model/utils/packing.py b/llava/model/utils/packing.py
index 4fc8908b..ef207130 100644
--- a/llava/model/utils/packing.py
+++ b/llava/model/utils/packing.py
@@ -33,3 +33,5 @@ def patch(model: nn.Module) -> None:
         m._get_unpad_data = _get_unpad_data
     else:
         transformers.modeling_flash_attention_utils._get_unpad_data = _get_unpad_data
+
+
diff --git a/llava/model/utils/utils.py b/llava/model/utils/utils.py
index d7899275..5a931eb5 100755
--- a/llava/model/utils/utils.py
+++ b/llava/model/utils/utils.py
@@ -172,3 +172,5 @@ def auto_upgrade(config):
         else:
             print("Checkpoint upgrade aborted.")
             exit(1)
+
+
diff --git a/llava/train/__init__.py b/llava/train/__init__.py
index e69de29b..139597f9 100755
--- a/llava/train/__init__.py
+++ b/llava/train/__init__.py
@@ -0,0 +1,2 @@
+
+
diff --git a/llava/train/args.py b/llava/train/args.py
index 47541733..07f67392 100755
--- a/llava/train/args.py
+++ b/llava/train/args.py
@@ -265,3 +265,5 @@ class TrainingArguments(transformers.TrainingArguments):
         default=False,
         metadata={"help": "Whether enter debug mode."},
     )
+
+
diff --git a/llava/train/callbacks/autoresume_callback.py b/llava/train/callbacks/autoresume_callback.py
index dd357d5c..0177ee96 100755
--- a/llava/train/callbacks/autoresume_callback.py
+++ b/llava/train/callbacks/autoresume_callback.py
@@ -60,3 +60,5 @@ def on_step_end(self, args, state, control, **kwargs):
                         AutoResume.request_resume()
                 control.should_training_stop = True
                 control.should_save = True
+
+
diff --git a/llava/train/deepspeed_replace/runtime/zero/mics.py b/llava/train/deepspeed_replace/runtime/zero/mics.py
index f48880ee..0587bf12 100644
--- a/llava/train/deepspeed_replace/runtime/zero/mics.py
+++ b/llava/train/deepspeed_replace/runtime/zero/mics.py
@@ -543,3 +543,5 @@ def load_state_dict(
         partition group we can call the load_state_dict logic from ZeRO-3.
         """
         super().load_state_dict(state_dict_list, load_optimizer_states, load_from_fp32_weights, checkpoint_folder)
+
+
diff --git a/llava/train/llava_trainer.py b/llava/train/llava_trainer.py
index a2eaf8aa..d1527faa 100755
--- a/llava/train/llava_trainer.py
+++ b/llava/train/llava_trainer.py
@@ -853,3 +853,5 @@ def log(self, logs: Dict[str, float]) -> None:
                     json.dump(self.state.log_history, f, indent=4)
 
         self.control = self.callback_handler.on_log(self.args, self.state, self.control, logs)
+
+
diff --git a/llava/train/sequence_parallel/__init__.py b/llava/train/sequence_parallel/__init__.py
index 960a2ba1..d80f8a89 100755
--- a/llava/train/sequence_parallel/__init__.py
+++ b/llava/train/sequence_parallel/__init__.py
@@ -1,2 +1,4 @@
 from .globals import get_pg_manager, set_pg_manager
 from .input_utils import extract_local_from_list, extract_local_input_ids, extract_local_position_ids
+
+
diff --git a/llava/train/sequence_parallel/all_to_all.py b/llava/train/sequence_parallel/all_to_all.py
index 96868035..b59a69d7 100755
--- a/llava/train/sequence_parallel/all_to_all.py
+++ b/llava/train/sequence_parallel/all_to_all.py
@@ -284,3 +284,5 @@ def forward(ctx: Any, group: dist.ProcessGroup, input: Any) -> Tensor:
     def backward(ctx: Any, grad_output: Tensor) -> Tuple[None, Tensor]:
         (tensor,) = ctx.saved_tensors
         return None, (None, tensor)
+
+
diff --git a/llava/train/sequence_parallel/globals.py b/llava/train/sequence_parallel/globals.py
index 9bfcb2aa..9854cf92 100755
--- a/llava/train/sequence_parallel/globals.py
+++ b/llava/train/sequence_parallel/globals.py
@@ -266,3 +266,5 @@ def get_data_parallel_size():
 def get_data_parallel_rank():
     """Get the rank of this process in the data parallel group the caller rank belongs to."""
     return PROCESS_GROUP_MANAGER.dp_rank
+
+
diff --git a/llava/train/sequence_parallel/hybrid_attn.py b/llava/train/sequence_parallel/hybrid_attn.py
index 39f08ccb..a0ce9daf 100755
--- a/llava/train/sequence_parallel/hybrid_attn.py
+++ b/llava/train/sequence_parallel/hybrid_attn.py
@@ -458,3 +458,5 @@ def forward(
 
     def backward(self, *args, **kwargs):
         raise RuntimeError("Backward computation is not allowed for AsyncHybridAttention.")
+
+
diff --git a/llava/train/sequence_parallel/input_utils.py b/llava/train/sequence_parallel/input_utils.py
index 294e837d..e61fef44 100755
--- a/llava/train/sequence_parallel/input_utils.py
+++ b/llava/train/sequence_parallel/input_utils.py
@@ -78,3 +78,5 @@ def extract_local_position_ids(input_ids, image_positions, image_ids, sp_rank, s
         return input_ids[start_position_idx:]
     else:
         return input_ids[start_position_idx:end_position_idx]
+
+
diff --git a/llava/train/sequence_parallel/monkey_patch.py b/llava/train/sequence_parallel/monkey_patch.py
index 993df46a..2560d97f 100755
--- a/llava/train/sequence_parallel/monkey_patch.py
+++ b/llava/train/sequence_parallel/monkey_patch.py
@@ -248,3 +248,5 @@ def _update_causal_mask(
     output_attentions: bool,
 ):
     return attention_mask
+
+
diff --git a/llava/train/sequence_parallel/ring/__init__.py b/llava/train/sequence_parallel/ring/__init__.py
index 841418b3..994fc22e 100755
--- a/llava/train/sequence_parallel/ring/__init__.py
+++ b/llava/train/sequence_parallel/ring/__init__.py
@@ -18,3 +18,5 @@
     zigzag_ring_flash_attn_varlen_func,
     zigzag_ring_flash_attn_varlen_qkvpacked_func,
 )
+
+
diff --git a/llava/train/sequence_parallel/ring/ring_flash_attn.py b/llava/train/sequence_parallel/ring/ring_flash_attn.py
index 16140dd7..a6b5f97f 100755
--- a/llava/train/sequence_parallel/ring/ring_flash_attn.py
+++ b/llava/train/sequence_parallel/ring/ring_flash_attn.py
@@ -298,3 +298,5 @@ def ring_flash_attn_func(
         return_attn_probs,
         group,
     )
+
+
diff --git a/llava/train/sequence_parallel/ring/ring_flash_attn_varlen.py b/llava/train/sequence_parallel/ring/ring_flash_attn_varlen.py
index 633e1819..d312347f 100755
--- a/llava/train/sequence_parallel/ring/ring_flash_attn_varlen.py
+++ b/llava/train/sequence_parallel/ring/ring_flash_attn_varlen.py
@@ -335,3 +335,5 @@ def ring_flash_attn_varlen_func(
         return_attn_probs,
         group,
     )
+
+
diff --git a/llava/train/sequence_parallel/ring/stripe_flash_attn.py b/llava/train/sequence_parallel/ring/stripe_flash_attn.py
index 6e0e8f27..6412ff52 100755
--- a/llava/train/sequence_parallel/ring/stripe_flash_attn.py
+++ b/llava/train/sequence_parallel/ring/stripe_flash_attn.py
@@ -342,3 +342,5 @@ def stripe_flash_attn_func(
         return_attn_probs,
         group,
     )
+
+
diff --git a/llava/train/sequence_parallel/ring/triton_utils.py b/llava/train/sequence_parallel/ring/triton_utils.py
index ca3aaad0..b11ade1d 100755
--- a/llava/train/sequence_parallel/ring/triton_utils.py
+++ b/llava/train/sequence_parallel/ring/triton_utils.py
@@ -154,3 +154,5 @@ def unflatten_varlen_lse(lse, cu_seqlens, max_seqlen: int):
             BLOCK_M,
         )
     return output
+
+
diff --git a/llava/train/sequence_parallel/ring/utils.py b/llava/train/sequence_parallel/ring/utils.py
index bf667506..8ba99574 100755
--- a/llava/train/sequence_parallel/ring/utils.py
+++ b/llava/train/sequence_parallel/ring/utils.py
@@ -122,3 +122,5 @@ def wait(self):
             req.wait()
         self._reqs = None
         self._ops = []
+
+
diff --git a/llava/train/sequence_parallel/ring/zigzag_ring_flash_attn.py b/llava/train/sequence_parallel/ring/zigzag_ring_flash_attn.py
index d3451313..1bb72efc 100755
--- a/llava/train/sequence_parallel/ring/zigzag_ring_flash_attn.py
+++ b/llava/train/sequence_parallel/ring/zigzag_ring_flash_attn.py
@@ -341,3 +341,5 @@ def zigzag_ring_flash_attn_func(
         return_attn_probs,
         group,
     )
+
+
diff --git a/llava/train/sequence_parallel/ring/zigzag_ring_flash_attn_varlen.py b/llava/train/sequence_parallel/ring/zigzag_ring_flash_attn_varlen.py
index f20d37cc..364f3864 100755
--- a/llava/train/sequence_parallel/ring/zigzag_ring_flash_attn_varlen.py
+++ b/llava/train/sequence_parallel/ring/zigzag_ring_flash_attn_varlen.py
@@ -459,3 +459,5 @@ def zigzag_ring_flash_attn_varlen_func(
         return_attn_probs,
         group,
     )
+
+
diff --git a/llava/train/sequence_parallel/ulysses_attn.py b/llava/train/sequence_parallel/ulysses_attn.py
index 00fa9422..d1f0d34b 100755
--- a/llava/train/sequence_parallel/ulysses_attn.py
+++ b/llava/train/sequence_parallel/ulysses_attn.py
@@ -229,3 +229,5 @@ def forward(
 
         # out e.g., [s/p::h]
         return output
+
+
diff --git a/llava/train/slurm_utils.py b/llava/train/slurm_utils.py
index 9af897a9..5b4b00ec 100755
--- a/llava/train/slurm_utils.py
+++ b/llava/train/slurm_utils.py
@@ -109,3 +109,5 @@ def on_step_end(self, args, state, control, model, **kwargs):
     def on_train_end(self, args, state, control, **kwargs):
         if state.global_step < state.max_steps:
             exit(124)
+
+
diff --git a/llava/train/train.py b/llava/train/train.py
index 2d1daf7c..eaf229c5 100755
--- a/llava/train/train.py
+++ b/llava/train/train.py
@@ -831,3 +831,5 @@ def make_inputs_require_grad(module, input, output):
 
 if __name__ == "__main__":
     train()
+
+
diff --git a/llava/train/train_hybrid.py b/llava/train/train_hybrid.py
index 60c2c6bb..6b25954c 100755
--- a/llava/train/train_hybrid.py
+++ b/llava/train/train_hybrid.py
@@ -42,3 +42,5 @@ def __iter__(self):
         mock.patch("accelerate.data_loader.BatchSamplerShard.__iter__", new=__iter__),
     ):
         train()
+
+
diff --git a/llava/train/train_llm_to_long.py b/llava/train/train_llm_to_long.py
index ab1c1158..9158b2c3 100644
--- a/llava/train/train_llm_to_long.py
+++ b/llava/train/train_llm_to_long.py
@@ -392,3 +392,5 @@ def forward_llm(
 
 if __name__ == "__main__":
     train()
+
+
diff --git a/llava/train/train_ln.py b/llava/train/train_ln.py
index 6a38305a..5bdf5cdb 100755
--- a/llava/train/train_ln.py
+++ b/llava/train/train_ln.py
@@ -863,3 +863,5 @@ def make_inputs_require_grad(module, input, output):
 
 if __name__ == "__main__":
     train()
+
+
diff --git a/llava/train/train_mem.py b/llava/train/train_mem.py
index 7fcca57f..630ec543 100755
--- a/llava/train/train_mem.py
+++ b/llava/train/train_mem.py
@@ -47,3 +47,5 @@ def __iter__(self):
     ):
         set_timer()
         train()
+
+
diff --git a/llava/train/train_mem_ln.py b/llava/train/train_mem_ln.py
index 8fb30127..fc3cdef5 100755
--- a/llava/train/train_mem_ln.py
+++ b/llava/train/train_mem_ln.py
@@ -47,3 +47,5 @@ def __iter__(self):
     ):
         set_timer()
         train()
+
+
diff --git a/llava/train/transformer_normalize_monkey_patch.py b/llava/train/transformer_normalize_monkey_patch.py
index 3003b541..f11a7122 100755
--- a/llava/train/transformer_normalize_monkey_patch.py
+++ b/llava/train/transformer_normalize_monkey_patch.py
@@ -295,3 +295,5 @@ def compute_loss(self, model, inputs, return_outputs=False, num_items_in_batch=N
         loss = outputs["loss"] if isinstance(outputs, dict) else outputs[0]
 
     return (loss, outputs) if return_outputs else loss
+
+
diff --git a/llava/train/utils.py b/llava/train/utils.py
index cbd33b25..a2b5c051 100755
--- a/llava/train/utils.py
+++ b/llava/train/utils.py
@@ -282,3 +282,5 @@ def sp_loss_rescale(shift_labels, loss):
 #     # dist.all_reduce(global_active_sum, group=get_ulysses_sp_pg())
 #     # loss_weight = num_active_elements / global_active_sum * PROCESS_GROUP_MANAGER.sp_degree
 #     # return loss_weight
+
+
diff --git a/llava/trl/__init__.py b/llava/trl/__init__.py
index a8d6de2c..af1d4d34 100644
--- a/llava/trl/__init__.py
+++ b/llava/trl/__init__.py
@@ -41,3 +41,5 @@
         DefaultDDPOStableDiffusionPipeline,
     )
     from .trainer import DDPOConfig, DDPOTrainer
+
+
diff --git a/llava/trl/core.py b/llava/trl/core.py
index 00997e59..43eb1643 100644
--- a/llava/trl/core.py
+++ b/llava/trl/core.py
@@ -339,3 +339,5 @@ def randn_tensor(
         latents = torch.randn(shape, generator=generator, device=rand_device, dtype=dtype, layout=layout).to(device)
 
     return latents
+
+
diff --git a/llava/trl/environment/__init__.py b/llava/trl/environment/__init__.py
index ae1cda4e..1cc1dc90 100644
--- a/llava/trl/environment/__init__.py
+++ b/llava/trl/environment/__init__.py
@@ -1,3 +1,5 @@
 # flake8: noqa
 
 from .base_environment import TextEnvironment, TextHistory
+
+
diff --git a/llava/trl/environment/base_environment.py b/llava/trl/environment/base_environment.py
index f83a30bd..a5b39d9f 100644
--- a/llava/trl/environment/base_environment.py
+++ b/llava/trl/environment/base_environment.py
@@ -470,3 +470,5 @@ def _generate_batched(
                 outputs.append(output[:generated_tokens])
         self.tokenizer.padding_side = padding_side_default
         return outputs
+
+
diff --git a/llava/trl/extras/__init__.py b/llava/trl/extras/__init__.py
index 6b3035db..756cea0d 100644
--- a/llava/trl/extras/__init__.py
+++ b/llava/trl/extras/__init__.py
@@ -14,3 +14,5 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 from .best_of_n_sampler import BestOfNSampler
+
+
diff --git a/llava/trl/extras/best_of_n_sampler.py b/llava/trl/extras/best_of_n_sampler.py
index 80810433..3bd16525 100644
--- a/llava/trl/extras/best_of_n_sampler.py
+++ b/llava/trl/extras/best_of_n_sampler.py
@@ -131,3 +131,5 @@ def generate(
             result.append(output)
 
         return result
+
+
diff --git a/llava/trl/extras/dataset_formatting.py b/llava/trl/extras/dataset_formatting.py
index 7583ce21..05973639 100644
--- a/llava/trl/extras/dataset_formatting.py
+++ b/llava/trl/extras/dataset_formatting.py
@@ -101,3 +101,5 @@ def get_formatting_func_from_dataset(
             return instructions_formatting_function(tokenizer)
 
     return None
+
+
diff --git a/llava/trl/import_utils.py b/llava/trl/import_utils.py
index 40d603b4..604d546f 100644
--- a/llava/trl/import_utils.py
+++ b/llava/trl/import_utils.py
@@ -105,3 +105,5 @@ def is_npu_available() -> bool:
     import torch_npu  # noqa: F401
 
     return hasattr(torch, "npu") and torch.npu.is_available()
+
+
diff --git a/llava/trl/models/__init__.py b/llava/trl/models/__init__.py
index f3879d71..a78ca1d2 100644
--- a/llava/trl/models/__init__.py
+++ b/llava/trl/models/__init__.py
@@ -31,3 +31,5 @@
         DDPOStableDiffusionPipeline,
         DefaultDDPOStableDiffusionPipeline,
     )
+
+
diff --git a/llava/trl/models/modeling_base.py b/llava/trl/models/modeling_base.py
index 0a44a93c..26cec6bb 100644
--- a/llava/trl/models/modeling_base.py
+++ b/llava/trl/models/modeling_base.py
@@ -669,3 +669,5 @@ def create_reference_model(
         logging.warning("Pattern passed or found, but no layers matched in the model. Check for a typo.")
 
     return ref_model.eval()
+
+
diff --git a/llava/trl/models/modeling_sd_base.py b/llava/trl/models/modeling_sd_base.py
index 6e5d30d5..b7a8ce9e 100644
--- a/llava/trl/models/modeling_sd_base.py
+++ b/llava/trl/models/modeling_sd_base.py
@@ -642,3 +642,5 @@ def load_checkpoint(self, models, input_dir):
             del load_model
         else:
             raise ValueError(f"Unknown model type {type(models[0])}")
+
+
diff --git a/llava/trl/models/modeling_value_head.py b/llava/trl/models/modeling_value_head.py
index 2771cc6c..777fb502 100644
--- a/llava/trl/models/modeling_value_head.py
+++ b/llava/trl/models/modeling_value_head.py
@@ -427,3 +427,5 @@ def generate(self, *args, **kwargs):
         We call `generate` on the wrapped model.
         """
         return self.pretrained_model.generate(*args, **kwargs)
+
+
diff --git a/llava/trl/models/utils.py b/llava/trl/models/utils.py
index 87d7ef39..363895a7 100644
--- a/llava/trl/models/utils.py
+++ b/llava/trl/models/utils.py
@@ -99,3 +99,5 @@ def setup_chat_format(
         model.generation_config.pad_token_id = tokenizer.pad_token_id
 
     return model, tokenizer
+
+
diff --git a/llava/trl/trainer/__init__.py b/llava/trl/trainer/__init__.py
index da33ad23..a5c87a50 100644
--- a/llava/trl/trainer/__init__.py
+++ b/llava/trl/trainer/__init__.py
@@ -43,3 +43,5 @@
 from .reward_config import RewardConfig
 from .reward_trainer import RewardTrainer, compute_accuracy
 from .sft_trainer import SFTTrainer
+
+
diff --git a/llava/trl/trainer/base.py b/llava/trl/trainer/base.py
index f0314cb9..c7db1b75 100644
--- a/llava/trl/trainer/base.py
+++ b/llava/trl/trainer/base.py
@@ -44,3 +44,5 @@ def compute_rewards(self, *args):
 
     def _save_pretrained(self, save_directory):
         raise NotImplementedError("Not implemented")
+
+
diff --git a/llava/trl/trainer/ddpo_config.py b/llava/trl/trainer/ddpo_config.py
index 6d692287..f3b8423b 100644
--- a/llava/trl/trainer/ddpo_config.py
+++ b/llava/trl/trainer/ddpo_config.py
@@ -134,3 +134,5 @@ def __post_init__(self):
                 "You need to install bitsandbytes to use 8bit Adam. "
                 "You can install it with `pip install bitsandbytes`."
             )
+
+
diff --git a/llava/trl/trainer/ddpo_trainer.py b/llava/trl/trainer/ddpo_trainer.py
index 107b70d7..c68a02fe 100644
--- a/llava/trl/trainer/ddpo_trainer.py
+++ b/llava/trl/trainer/ddpo_trainer.py
@@ -626,3 +626,5 @@ def create_model_card(self, path: str, model_name: Optional[str] = "TRL DDPO Mod
     def _save_pretrained(self, save_directory):
         self.sd_pipeline.save_pretrained(save_directory)
         self.create_model_card(save_directory)
+
+
diff --git a/llava/trl/trainer/dpo_trainer.py b/llava/trl/trainer/dpo_trainer.py
index 9f6f230b..bd46fc15 100644
--- a/llava/trl/trainer/dpo_trainer.py
+++ b/llava/trl/trainer/dpo_trainer.py
@@ -1307,3 +1307,5 @@ def push_to_hub(self, commit_message: Optional[str] = "End of training", blockin
         kwargs = trl_sanitze_kwargs_for_tagging(model=self.model, tag_names=self._tag_names, kwargs=kwargs)
 
         return super().push_to_hub(commit_message=commit_message, blocking=blocking, **kwargs)
+
+
diff --git a/llava/trl/trainer/iterative_sft_trainer.py b/llava/trl/trainer/iterative_sft_trainer.py
index f25431d8..c33c3e40 100644
--- a/llava/trl/trainer/iterative_sft_trainer.py
+++ b/llava/trl/trainer/iterative_sft_trainer.py
@@ -364,3 +364,5 @@ def _maybe_log_save_evaluate(self):
                 self._globalstep_last_logged = self.state.global_step
 
                 self.log(logs)
+
+
diff --git a/llava/trl/trainer/model_config.py b/llava/trl/trainer/model_config.py
index 8ac4a9fa..50339e1f 100644
--- a/llava/trl/trainer/model_config.py
+++ b/llava/trl/trainer/model_config.py
@@ -98,3 +98,5 @@ def to_dict(self):
     def __post_init__(self):
         if self.load_in_8bit and self.load_in_4bit:
             raise ValueError("You can't use 8 bit and 4 bit precision at the same time")
+
+
diff --git a/llava/trl/trainer/ppo_config.py b/llava/trl/trainer/ppo_config.py
index 6564c383..e8565a4f 100644
--- a/llava/trl/trainer/ppo_config.py
+++ b/llava/trl/trainer/ppo_config.py
@@ -176,3 +176,5 @@ def to_dict(self):
         for key, value in self.__dict__.items():
             output_dict[key] = value
         return flatten_dict(output_dict)
+
+
diff --git a/llava/trl/trainer/ppo_trainer.py b/llava/trl/trainer/ppo_trainer.py
index e9873dcf..fb17b347 100644
--- a/llava/trl/trainer/ppo_trainer.py
+++ b/llava/trl/trainer/ppo_trainer.py
@@ -1452,3 +1452,5 @@ def _prepare_deepspeed(self, model: PreTrainedModelWrapper):
         model, *_ = deepspeed.initialize(model=model, config=config_kwargs)
         model.eval()
         return model
+
+
diff --git a/llava/trl/trainer/reward_config.py b/llava/trl/trainer/reward_config.py
index 32c7e264..2cea6b5a 100644
--- a/llava/trl/trainer/reward_config.py
+++ b/llava/trl/trainer/reward_config.py
@@ -36,3 +36,5 @@ class RewardConfig(TrainingArguments):
 
     max_length: Optional[int] = None
     """The maximum length of the sequences in the batch. This argument is required if you want to use the default data collator."""
+
+
diff --git a/llava/trl/trainer/reward_trainer.py b/llava/trl/trainer/reward_trainer.py
index a6048beb..f24e0e38 100644
--- a/llava/trl/trainer/reward_trainer.py
+++ b/llava/trl/trainer/reward_trainer.py
@@ -272,3 +272,5 @@ def prediction_step(
         labels = self._prepare_inputs(labels)
 
         return loss, logits, labels
+
+
diff --git a/llava/trl/trainer/sft_trainer.py b/llava/trl/trainer/sft_trainer.py
index be8c6b45..01dc9cbd 100644
--- a/llava/trl/trainer/sft_trainer.py
+++ b/llava/trl/trainer/sft_trainer.py
@@ -510,3 +510,5 @@ def _trl_activate_neftune(self, model):
         hook_handle = embeddings.register_forward_hook(neftune_post_forward_hook)
         self.neftune_hook_handle = hook_handle
         return model
+
+
diff --git a/llava/trl/trainer/utils.py b/llava/trl/trainer/utils.py
index ac43bc69..68f77890 100644
--- a/llava/trl/trainer/utils.py
+++ b/llava/trl/trainer/utils.py
@@ -722,3 +722,5 @@ def get_peft_config(model_config: ModelConfig) -> "Optional[PeftConfig]":
     )
 
     return peft_config
+
+
diff --git a/llava/utils/__init__.py b/llava/utils/__init__.py
index 16281fe0..d81436df 100644
--- a/llava/utils/__init__.py
+++ b/llava/utils/__init__.py
@@ -1 +1,3 @@
 from .utils import *
+
+
diff --git a/llava/utils/distributed.py b/llava/utils/distributed.py
index f4c855e8..92ebcd51 100644
--- a/llava/utils/distributed.py
+++ b/llava/utils/distributed.py
@@ -71,3 +71,5 @@ def all_gather(obj: Any) -> List[Any]:
     objs = [None for _ in range(size())]
     dist.all_gather_object(objs, obj)
     return objs
+
+
diff --git a/llava/utils/io.py b/llava/utils/io.py
index 23240c0a..03f4ea6c 100644
--- a/llava/utils/io.py
+++ b/llava/utils/io.py
@@ -169,3 +169,5 @@ def save(fpath: str, obj: Any, **kwargs) -> None:
             return
 
     raise NotImplementedError(f'"{fpath}" cannot be saved.')
+
+
diff --git a/llava/utils/logging.py b/llava/utils/logging.py
index fdabdece..7c967e38 100644
--- a/llava/utils/logging.py
+++ b/llava/utils/logging.py
@@ -15,3 +15,5 @@ def __get_logger() -> Logger:
 
 
 logger = __get_logger()
+
+
diff --git a/llava/utils/media.py b/llava/utils/media.py
index 7d41c4b6..ff877ab3 100644
--- a/llava/utils/media.py
+++ b/llava/utils/media.py
@@ -103,3 +103,5 @@ def extract_media(
                 raise ValueError(f"Unsupported prompt part type: {type(part)}")
         message["value"] = text
     return media
+
+
diff --git a/llava/utils/merge_lora_weights_and_save_hf_model.py b/llava/utils/merge_lora_weights_and_save_hf_model.py
index e80a3f09..a1d48f42 100644
--- a/llava/utils/merge_lora_weights_and_save_hf_model.py
+++ b/llava/utils/merge_lora_weights_and_save_hf_model.py
@@ -103,3 +103,5 @@ def main():
 
 if __name__ == "__main__":
     main()
+
+
diff --git a/llava/utils/tokenizer.py b/llava/utils/tokenizer.py
index a0d32697..daf1d98c 100644
--- a/llava/utils/tokenizer.py
+++ b/llava/utils/tokenizer.py
@@ -181,3 +181,5 @@ def infer_stop_tokens(tokenizer: transformers.PreTrainedTokenizer) -> List[str]:
             stop_token = tokenizer.decode(template[k + 1])
             stop_tokens.add(stop_token)
     return list(stop_tokens)
+
+
diff --git a/llava/utils/utils.py b/llava/utils/utils.py
index 151e6fbd..41a13294 100755
--- a/llava/utils/utils.py
+++ b/llava/utils/utils.py
@@ -31,3 +31,5 @@ def disable_torch_init():
 
     setattr(torch.nn.Linear, "reset_parameters", lambda self: None)
     setattr(torch.nn.LayerNorm, "reset_parameters", lambda self: None)
+
+
diff --git a/llava/wids/__init__.py b/llava/wids/__init__.py
index 6b9b3a40..a497b8f1 100755
--- a/llava/wids/__init__.py
+++ b/llava/wids/__init__.py
@@ -5,3 +5,5 @@
 # flake8: noqa
 
 from .wids import ChunkedSampler, DistributedChunkedSampler, DistributedLocalSampler, ShardedSampler, ShardListDataset
+
+
diff --git a/llava/wids/wids.py b/llava/wids/wids.py
index f34f5351..5d2ea264 100755
--- a/llava/wids/wids.py
+++ b/llava/wids/wids.py
@@ -823,3 +823,5 @@ def __iter__(self):
         # print("[SamplerIndices: ]", indices)
         assert len(indices) == self.num_samples
         return iter(indices)
+
+
diff --git a/llava/wids/wids_bench.py b/llava/wids/wids_bench.py
index 92044b69..fe1b06d0 100755
--- a/llava/wids/wids_bench.py
+++ b/llava/wids/wids_bench.py
@@ -45,3 +45,5 @@ def main_wds(args):
         main_wds(args)
     else:
         raise ValueError(f"Unknown command: {args.command}")
+
+
diff --git a/llava/wids/wids_cleanup.py b/llava/wids/wids_cleanup.py
index efddf294..d5e043b7 100755
--- a/llava/wids/wids_cleanup.py
+++ b/llava/wids/wids_cleanup.py
@@ -92,3 +92,5 @@ def cleanup_worker(every):
     p = multiprocessing.Process(target=cleanup_worker, args=(every,))
     p.start()
     return p
+
+
diff --git a/llava/wids/wids_dir.py b/llava/wids/wids_dir.py
index dd53dac7..a9777ab6 100755
--- a/llava/wids/wids_dir.py
+++ b/llava/wids/wids_dir.py
@@ -19,3 +19,5 @@ def __iter__(self):
         # rename file back to its original name or unlink it, according to strategy
         pass
 """
+
+
diff --git a/llava/wids/wids_dl.py b/llava/wids/wids_dl.py
index 307248f7..1b37dd11 100755
--- a/llava/wids/wids_dl.py
+++ b/llava/wids/wids_dl.py
@@ -154,3 +154,5 @@ def download_and_open(remote, local, mode="rb", handlers=default_cmds, verbose=F
             key = tuple(str(x) for x in [remote, local, mode, current_time])
             open_objects[key] = result
         return result
+
+
diff --git a/llava/wids/wids_index.py b/llava/wids/wids_index.py
index 8b522b58..60c40662 100755
--- a/llava/wids/wids_index.py
+++ b/llava/wids/wids_index.py
@@ -317,3 +317,5 @@ def main():
 
 if __name__ == "__main__":
     main()
+
+
diff --git a/llava/wids/wids_lru.py b/llava/wids/wids_lru.py
index b288094f..89d20f89 100755
--- a/llava/wids/wids_lru.py
+++ b/llava/wids/wids_lru.py
@@ -62,3 +62,5 @@ def clear(self):
 
     def __del__(self):
         self.clear()
+
+
diff --git a/llava/wids/wids_mmtar.py b/llava/wids/wids_mmtar.py
index d8a3590d..d239c329 100755
--- a/llava/wids/wids_mmtar.py
+++ b/llava/wids/wids_mmtar.py
@@ -149,3 +149,5 @@ def keep_while_reading(fname, fd, phase, delay=0.0):
             pass
     else:
         raise ValueError(f"Unknown phase {phase}")
+
+
diff --git a/llava/wids/wids_specs.py b/llava/wids/wids_specs.py
index 7a3aa11c..adffd141 100755
--- a/llava/wids/wids_specs.py
+++ b/llava/wids/wids_specs.py
@@ -173,3 +173,5 @@ def load_dsdesc_and_resolve(source, *, options=None, base=None):
         options = {}
     dsdesc = load_remote_dsdesc_raw(source)
     return resolve_dsdesc(dsdesc, base=base, options=options)
+
+
diff --git a/llava/wids/wids_tar.py b/llava/wids/wids_tar.py
index 7a62fd0e..e8649994 100755
--- a/llava/wids/wids_tar.py
+++ b/llava/wids/wids_tar.py
@@ -79,3 +79,5 @@ def get_file(self, i):
     def close(self):
         # Close the tar file
         self.tar_file.close()
+
+
diff --git a/server.py b/server.py
index 2b3f1e28..24201b31 100644
--- a/server.py
+++ b/server.py
@@ -288,3 +288,5 @@ def chunk_generator():
     app.args = parser.parse_args()
 
     uvicorn.run(app, host=app.args.host, port=app.args.port, workers=app.args.workers)
+
+